The artificial intelligence (AI) Generative Pre-trained Transformer 3 (GPT-3) can write texts on any topic. OpenAI, the organisation that developed and released it as a beta version in June 2020, describes it as a general-purpose application for creating text, ‘allowing users to try it on virtually any English language task’. GPT-3 is a scaling up, by two orders of magnitude, of the previous model released by OpenAI, making it ‘the most powerful natural language processing (NLP) application available today’. 

The promises are greater accuracy and an improved ability to transfer things learned in one context to a different context. Overall, GPT-3 can mimic a variety of styles and genres, and in doing so, return texts that look very much like having been written by a human. The Guardian recently used it to write an article. So, what does this mean for diplomats whose daily work is steeped in the art and craft of language? 

Automated diplomacy?

When thinking through the use of AI for specific tasks and within specific professions, it is useful to distinguish between augmentation and automation. Augmentation describes a situation where parts of a task are taken over by a machine. Automation means that the whole process is taken over by a machine with extremely minimal, if any, human intervention. What can GPT-3 deliver in terms of augmented and automated diplomacy? 

Augmentation: Efficiency tools

OpenAI’s website includes a number of use cases that are also applicable to the work of diplomats. First, the company CaseText uses GPT-3 to search through legal documents and to facilitate litigations and presentations by lawyers. Similar applications in the area of international law are not hard to imagine, and have indeed already been suggested and tested (the Cognitive Trade Advisor is an example). Second, productivity tools that lead to better decisions could also be applied in the field of diplomatic practice. Third, ‘comprehension tools’, that provide quick summaries of long texts, might also eventually aid the work of diplomats. As these tools become more widely available and used, it is not far-fetched to suggest that diplomats will use them in their daily work, either as off-the-shelf productivity tools or as custom-build systems that take the specifics of the work of diplomats into account. With GPT-3 becoming available beyond the beta version, developing custom applications should move within easy reach. It’s also worth pointing out that the tools described here are nothing new, the difference being that GPT-3 is the latest and most powerful NLP tool available today. 

The promise associated with use cases like these is greater efficiency and productivity. While this resonates well in a business context, it resonates less when it comes to diplomatic practice. To be clear, ministries of foreign affairs are under budgetary constraints and have an obligation to use public money responsibly. It can also be an advantage to be faster and more efficient when doing research in preparation for a negotiation. However, finding an agreement or being successful in negotiating texts cannot be measured by these efficiency metrics. While greater efficiency can be an advantage for negotiators and can level the planning field for small and developing states, it does not win you the overall ‘battle’. 

Automation: Diplomatic writing tasks

GPT-3 delivers some interesting results on the basis of an initial short piece of text submitted to the system. It matches the tone and style and returns a text that is, more often than not, understandable and reasonable. More importantly, it is hard, if not impossible, to distinguish that text from a piece written by a human being. 

Therefore, we can assume that the system will be able to match the tone and style of a typical diplomatic speech, for example, those delivered at the opening of the UN General Assembly each year. It is also feasible that it will match certain positions and interests based on the initial short text submitted to it. If you give the system a speech by Prime Minister of New Zealand Jacinda Ardern, it will very likely return a text that believably sounds like a speech by her. If you give the system a speech by US President Donald Trump, it will very likely return a text that believably sounds like a speech by him. 

While such a text might be interesting as an initial suggestion or a general template, it will need a lot of editing and rewriting. Although we were not able to test GPT-3 ourselves, we assume that the text, also passable as having been written by a human being, will still miss the mark in the context of diplomatic practice. The following aspects are very likely missing: overall coherence; references to specific examples that are most useful in this context; references to historic moments important for an occasion; and an understanding of the relations between countries and how they should be reflected, often implicitly, in specific parts of the speech. 

The explanation for these doubts and potential shortcomings is simple: GPT-3 operates by mapping relationships between words without having an understanding of the meaning of the words. It’s great at predicting the next word in a sentence, but lacks understanding of the overall context. This explains the statement from Open AI that GPT-3’s ‘success generally varies depending on how complex the task is’. For these more complex tasks, human editors and writers are needed. For example, it’s also worth noting that, according to the editor’s note accompanying the Guardian article mentioned above, the article was a piece of augmented, not automated, journalism. Journalists selected and rearranged passages, and the article went through the usual editing process. An opinion piece also published in the Guardian suggested that 90% of the text generated by GPT-3 was discarded before editing. 

This is not to take away from the fact that GPT-3 is a huge accomplishment and a big step for these types of language processing AIs. It might serve as a way of making speech-writing quicker by already providing templates and useful suggestions. In this sense, it could work much like the autocomplete function in e-mail services and word processors. This brings us back to the automation-vs-augmentation question, and the, perhaps, reassuring knowledge that neither diplomats nor human speech-writers are likely to be replaced anytime soon. 

The way forward?

Without having tested GPT-3 ourselves, we cannot be sure, but the hunch is that more specialised systems are needed in the area of diplomacy. In a paper released by the mothers and fathers of GPT-3, it is suggested that relying on a more-text-more-computing-power approach will eventually come up against limits. With such an approach, the system becomes better and better at predicting the word most likely to appear next in a sentence. It does not, however, become better at keeping the next sentence or the text as a whole ‘in mind’ (for a detailed discussion of this point, see this article on GPT-3). For that, a different approach is needed. 

At DiploFoundation, as part of our AI humAInism project, we have experimented with how this different approach could look like in the field of diplomacy. Our own Speech Generator is meant as an illustration of what can be done and how it can be done. Diplomats working in the field of digital policy and cybersecurity will find it particularly interesting to experiment with. The Speech Generator allows for selecting an opinion on various key topics on the basis of which a speech is generated. 

In contrast to applications like GPT-3, we tried to mimic the human process of writing a speech by using smaller algorithms trained for specific tasks, such as an algorithm for  finding keywords and phrases (‘underlining’), an algorithm for recommending paragraphs on a specific topic, an algorithm for summarising paragraphs, etc. As our developer Jovan Njegic would say, ‘in this way, we try to form a system of interconnected algorithms, which imitate not the results of the writing process, but the human process of reasoning during speech-writing’. This also means that if a result is not appropriate, the user can go back and tweak the process. Our speech generator is an illustration, not a fully fledged application for diplomats, but it might just point us in the right future direction. 

On 21 September, DiploFoundation launched the humAInism Speech Generator as part of its humAInism project. By combining artificial intelligence (AI) algorithms and the expertise of Diplo’s cybersecurity team, this tool is meant to help diplomats and practitioners write speeches on the topic of cybersecurity. 

Given the research nature of the project, the main purpose of the generator was to explore various new AI technologies and examine their useability in the field of diplomacy. For this purpose, we used several state-of-the-art algorithms for the generator, with three main purposes. 

1. Semantic similarity search: Finding sentences with similar semantics from DiploFoundation’s corpuses of books and transcripts.
2. Generation of long-form answers: Given a question, the algorithm finds relevant paragraphs from Diplo’s corpuses of books and transcripts, and generates new paragraphs with explanatory answers.
3. Text generation: The algorithm is fine-tuned on diplomatic texts, and is used for the generation of new texts.

1. Semantic similarity search

We use the DistilBERT language representation model to encode sentences into 512-dimensional vectors. After that, the approximate nearest neighbor search algorithm is used to compare vectors and calculate their similarity score according to their angular distance. For this purpose, we implemented the technology listed below. 

1.1. DistilBERT model

DistilBERT is a transformers model, smaller and faster than BERT (Bidirectional Encoder Representations from Transformers), which was pretrained on the same corpuses in a self-supervised fashion using the BERT base model as a teacher. This means it was pretrained on raw texts only, with no humans labelling them in any way (which is why it can use a lot of publicly available data), and through an automatic process generate inputs and labels from those texts using the BERT base model. More precisely, it was pretrained with three objectives. 

• Distillation loss: The model was trained to return the same probabilities as the BERT base model.
• Masked language modeling (MLM): This is part of the original training loss of the BERT base model. When taking a sentence, the model randomly masks 15% of the words in the input, then runs the entire masked sentence through the model, and predicts the masked words. This is different from the traditional recurrent neural networks (RNNs) that usually see the words one after the other, or from the autoregressive models, like the generative pretrained transformer (GPT), which internally mask future tokens; it allows the model to learn a bidirectional representation of a sentence.
• Cosine embedding loss: The model was also trained to generate hidden states as close as possible to the BERT base model.

In this way, the model learns the same inner representation of the English language as its teacher model, while being faster for inference and downstream tasks.

Reference: Sanh V et al. (2019) DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter. arXiv, 1 March. Available at https://arxiv.org/abs/1910.01108 [accessed 20 September 2020].

1.2. Approximate Nearest Neighbors Oh Yeah (Annoy) search algorithm

Annoy is a C++ library with Python bindings which searches for points in space that are close to a given query point. It also creates large read-only file-based data structures that are mapped into the memory so that many processes may share the same data.

Reference: The Annoy Python module on GitHub. Available at https://github.com/spotify/annoy [accessed 20 September 2020].

2. Generation of long-form answers

For this task, we used models that were pretrained on Wikipedia (Wiki-40B) and the Explain Like I’m Five (ELI5) questions datasets. We applied models on our custom Diplo dataset, consisting of Diplo books and Internet Governance Forum (IGF) transcripts. The process of generating answers is done in two stages. 

1. At the retrieval stage, the pretrained custom-made embedder is used to project a BERT 512 embedding vector to a 128-dimensional space in a way that the dot inner product of the projection of the question vector and a projection of answer vector should be higher than the dot inner products of the projection of the question vector and a projection of any other answer vector. Document retrieval is conducted by the Max Inner Product Search (MIPS) of dense 128 embeddings with Faiss.
2. At the generation stage, the pretrained BART sequence-to-sequence model is used for generating answers.

The applied algorithms are listed below. 

2.1. BERT 

The language representation model BERT is short for ‘Bidirectional Encoder Representations from Transformers’. Unlike recent language representation models, BERT is designed to pretrain deep bidirectional representations from an unlabeled text by jointly conditioning both left and right contexts in all layers. As a result, the pretrained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks, such as question answering (QA) and language inference, without substantial task-specific architecture modifications.

Reference: Devlin J et al. (2018) BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. arXiv, 11 October. Available at https://arxiv.org/abs/1810.04805 [accessed 20 September 2020]. 

2.2. Faiss search algorithm

The Facebook AI Similarity Search (Faiss) is a library for efficient similarity search and the clustering of dense vectors. It contains algorithms that search in sets of vectors of any size, up to those that possibly do not fit into the random-access memory (RAM). It also contains a supporting code for evaluation and parameter tuning. Faiss is written in C++, with complete wrappers for Python/NumPy. Some of the most useful algorithms are implemented on the graphics processing unit (GPU). Faiss was developed by Facebook Artificial Intelligence Research (FAIR).

Reference: Faiss on GitHub. Available at https://github.com/facebookresearch/faiss/wiki [accessed 20 September 2020].

2.3. BART

BART is a denoising autoencoder for pretraining sequence-to-sequence models. It is trained by: (1) corrupting text with an arbitrary noising function, and (2) teaching a model to reconstruct the original text. It uses a standard tranformer-based neural machine translation architecture which, despite its simplicity, can be seen as generalising BERT (due to the bidirectional encoder), GPT (with a left-to-right decoder), and many other more recent pretraining schemes. BART is particularly effective when fine-tuned for generating text, but it also works well for comprehension tasks. It matches the performance of RoBERTa, with comparable training resources on GLUE (General Language Understanding Evaluation) and SQuAD (Stanford Question Answering Dataset); and achieves new state-of-the-art results on a range of abstract dialogues, question answering, and summarisation tasks, with gains of up to 6 ROUGE (Recall-Oriented Understudy for Gisting Evaluation). BART also provides a 1.1 BLEU (bilingual evaluation understudy) increase over a back-translation system for machine translation, with only target language pretraining.

Reference: Lewis M et al. (2019) BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension. arXiv, 29 October. Available at https://arxiv.org/abs/1910.13461 [accessed 20 September 2020].

3. Text generation

For the task of generating introductory sentences, we used a pretrained GPT-2 algorithm, which we fine-tuned on a dataset generated from the first three sentences of the UN General Debates Dataset (UN General Debates). 

Elements of the applied algorithm are listed below. 

3.1. GPT-2 

‘GPT-2 is a large transformer-based language model trained using the simple task of predicting the next word in 40GB of high-quality text from the internet. This simple objective proves sufficient to train the model to learn a variety of tasks due to the diversity of the dataset. In addition to its incredible language generation capabilities, it is also capable of performing tasks like question answering, reading comprehension, summarisation, and translation. While GPT-2 does not beat the state-of-the-art in these tasks, its performance is impressive nonetheless considering that the model learns these tasks from raw text only.’ (Rajapakse, 2020) 

Reference: 

Radford A et al. Language Models are Unsupervised Multitask Learners. OpenAI. Available at https://arxiv.org/abs/1910.13461 [accessed 20 September 2020].

Rajapakse T (2020) Learning to Write: Language Generation With GPT-2. Medium, 27 April. Available at https://medium.com/swlh/learning-to-write-language-generation-with-gpt-2-2a13fa249024 [accessed 20 September 2020].

As we face unprecedented change and an uncertain future, it is the right moment to revisit the fundamentals upon which our society is built. A full examination of the values, ideas, and concepts that have worked in our society is necessary to guide what will work as we grapple with accelerating modernity. With technologies such as artificial intelligence (AI), advanced genetic modification, and automated weapons all quickly becoming a reality, our humanity will be challenged like never before. The search for solutions should start by going ‘back to fundamentals’, as was done during last week’s discussion in Zurich organised by the European Forum Alpbach and the Dezentrum and Foraus think tanks. The following questions echoed:

Can we simply go back to the fundamentals and build a better society or do we need to revisit and perhaps revise these very fundamentals? What are the fundamentals of society? How far back in history should we go to best determine the fundamentals of modern society?

The search for societal fundamentals brings us as far back as the ‘Axial Age’, when our ancestors began grasping their destiny via spiritual transcendence and rational agency. In the span of five centuries (500 BC – 0 AD), Hinduism, Buddhism, and Jainism were born in India while Taoism and Confucianism took hold in China. Philosophers Socrates, Plato, and Aristotle emerged in Greece and the Second Temple, Judaism, and Christianity came to be in the Middle East. With Islam, which emerged a few centuries later, humanity developed a ‘societal software’ that still operates today.

Fast forward to the Enlightenment when Descartes, Hume, Rousseau, Voltaire, Kant, and other thinkers put humans and rationality at the centre of societal developments. The two pillars of the Enlightenment – modernity and humanity – have shaped our world to this day.

Modernity has human rationality and progress in its core. Modernity gave science and technology the ability to truly grow. Industries flourished. Societies developed. Endowed with longevity, human life became less dangerous and more enjoyable. Modernity optimised the use of time and resources. 

Humanity, the other pillar of the Enlightenment, put humans at the centre of society. Its key tenants are respect for human life and dignity, the realisation of human potential, and the individual right to make personal, economic and political choices. Although they originated during the Enlightenment, humanity’s values were eventually codified in core international documents, such as the Universal Declaration on Human Rights and the UN Charter by the mid-20th century.

Over the past few centuries, modernity and humanity reinforced each other in a virtuous cycle of sorts. Advancements in science and technology helped attain the emancipation of millions worldwide. The free and more educated mind breathed creativity and ingenuity into science and technology. The Enlightenment formula seemed to work. 

However, in the last decade, tensions between modernity and humanity have started to emerge with the fast growth of digital technology, and, in particular, AI. This has made us harbour questions about our future. 

Will advanced technology reduce the space for human agency and, ultimately, our right to make personal, political, and economic choices?

From time immemorial, we have been making choices using our brains (logos), our hearts (ethos), and our gut (pathos). Those choices, good or bad, were ours, and ours alone. Suddenly, machines became capable of making more optimal and informed choices. They started gathering enormous amounts of data and more importantly, they started gathering data about us – what we like, what we search for, what we purchase, where we go, and how we get there. The algorithms behind the machines came to know us better than we know ourselves. 

As machines start to gradually replace our human agency to choose; from helping us identify our lifelong partner, to showing us what item we should purchase next. We need to ask ourselves whether we will still be able to resist this advice if we want to. While it may be tempting to allow AI to choose for us, a blanket reliance on it can have far-reaching consequences on our society, economy, and politics. 

To solve this growing dilemma, we will need to revisit the interplay between modernity and humanity. Will modernity and humanity continue to reinforce each other, or will modernity, driven by science technology, stifle humanity? Should we safeguard our right to human imperfection, especially, in situations where our abilities are no match for AI and machines? 

These and other questions will remain with us in the coming years as we discuss a new social contract that can capture our shared understanding on the future of humanity. At worst, we need to avoid the autoimmune trap where modernity harms our core humanity. At best, we need to find new ways how modernity and humanity can continue reinforcing each other.

If you’ve recognised the sound in the audio file above, you are old enough to remember the birth of the Internet as we know it.

In 2000, the two of us wrote a book titled the Internet Guide for Diplomats. At the time, it was the only book to explain the opportunities and changes that the Internet revolution was bringing to diplomacy. 

Our idea was simple. We wanted to provide a practical handbook for diplomats to make the best use of what was a new powerful tool. IT literacy among diplomats was almost non-existent. The changes that the Internet brought to the diplomatic professional life took place too quickly to be metabolised. Therefore, there was a need for training in new skills and new tools. The powerful new tool was now at the diplomat’s complete disposal.

It was also a time when a limited number of organisations and institutions – both local and global – had a fully functional website. E-mail was not yet part of diplomats’ daily routine.

Fast forward by 20 years

Today, every organisation (and ministry) has its communication strategy that relies heavily on the Internet and social media. Limitations in the use of videos, due to bandwidth limitations, have been overcome thanks to fast broadband and new devices which have made it easy to produce and broadcast videos.

How many of the concepts that we covered two decades ago in our book are still valid today? While the technology and telecommunications landscapes have changed a lot, some principles are still part of today’s challenges. 

Over time, data replaced information in the digital lingo, but issues of data or information management and governance remain very similar. For example, what we referred to back then as ‘how to evaluate information’ is what we are tackling today as data science. Handling misinformation has evolved into dealing with fake news. The question ‘who governs the Internet?’ is still relevant and has received no valid answer. Internet governance is still one of the issues that the international community has been heavily discussing, without being able to find sufficient common ground. 

Among diplomats, tech-literacy and knowledge of all things digital has certainly improved. Yet, the developments in technology and communications are evolving very quickly. Diplomatic expertise in social media and digital marketing, two essential elements for effective public communication, still need improvement. New skills for chairing online meetings or using artificial intelligence (AI) for diplomatic reporting are needed more and more.

Diplomatic training in these specific areas, therefore, need to be developed, and programmes of diplomatic training institutions need to be adjusted to the current needs. Reassuringly, the use of hypertext, Internet conferencing, discussion groups, and cognitive maps – already present 20 years ago – still provide the most solid basis for online learning. 

Ahead of us, the pendulum between continuity and change will keep its rhythmic swing. We will see continuity through the core functions of diplomacy – a tool for the peaceful solution of conflicts and for the promotion of national interests. We will also see many changes triggered by new tools (from AI to augmented reality) for negotiations and other diplomatic activities. Are we ready?

More resources

• Visit our portals dedicated to e-diplomacy and ConfTech Lab for our latest research. 
• Apply for our Master’s in Contemporary Diplomacy (now, completely online!) which focuses on the future of diplomacy.
• Take a short course and sharpen your thinking on e-diplomacy, negotiations, and online meetings. 
• For more information, and to share suggestions on how to advance the discussion on the future of diplomacy, drop us a line.

On June 11, UN Secretary General António Guterres launched the Roadmap for Digital Cooperation based on the report of the High-level Panel on Digital Cooperation. The launch was accompanied by a series of discussions on the Roadmap’s key recommendations.

In this blog post, I want to zoom in on the key points related to artificial intelligence (AI) as they are presented in the Roadmap, and as they were reflected on in the discussions that ensued. I will highlight five key points from the Roadmap, five main impressions from the Roadmap’s launch, and four examples of the initial reactions and commentary.

Key points from the Roadmap

Under the heading ‘Human Rights and Human Agency’, the Roadmap’s Recommendation 3c directly addresses AI.

First, the Roadmap clearly recognises the advances made in AI and its associated challenges and opportunities. It highlights that AI applications are already ubiquitous and that substantial added value to global markets is to be expected in the coming years. At the same time, the Roadmap is clear that AI can ‘significantly compromise the safety and agency of users worldwide’.

Second, on a normative level, the Roadmap emphasises that AI, in particular facial recognition, should not be used to:

• ‘erode human rights’;
• ‘deepen inequality’; or
• ‘exacerbate existing discrimination’.

The Roadmap states that AI should be put to work towards achieving the sustainable development goals (SDGs) and that it ‘must not be used to erode human rights’. It also reiterates the call of the UN secretary general for a global ban on lethal autonomous weapons (LAWs).

Third, the Roadmap identifies gaps in international co-ordination, co-operation, and governance when it comes to AI. More specifically, it highlights:

• the ‘lack of representation and inclusiveness in global discussions’;
• the ‘lack of overall coordination’ among AI initiatives; and
• the absence of a ‘common platform to bring these separate initiatives together’.

Furthermore, national governance and oversight ‘would benefit from additional capacity and expertise’.

Fourth, while AI is explicitly discussed in Recommendation 3c, other parts of the Roadmap are also relevant for AI, including the Roadmap’s recommendations on digital public goods (1b), digital capacity building (2), digital trust and security (4), and global digital co-operation (5). All can and should also be read with AI in mind. Further, the recommendations on digital human rights (3a and 3b) mention two points with direct implications for AI development and deployment: data protection and privacy; and surveillance technologies, including facial recognition. It is clear that the report’s relevance for the future of AI extends beyond Recommendation 3c.

Fifth, regarding the actions to be undertaken by the UN secretary general, establishing a multi-stakeholder coalition for digital inclusion and appointing a technology envoy in 2021, will have bearing on AI development and governance, while establishing the multi-stakeholder advisory body on global co-operation on AI will create an important inclusive venue for taking forward the discussions on AI.

Key points from the global launch event

After the June 11 launch event:

two dialogues accompanying the launch took place on June 12 and June 15. The discussion on June 15:

which I’m focusing on here, specifically addressed AI, but the topic was mentioned on several occasions during all three events.

First, in terms of applications of AI, the discussion was focused on the progress towards the SDGs and the responses to the COVID-19 crisis. UN Under-Secretary-General Fabrizio Hochschild, who guided the discussion, emphasised that the development towards reaching the SDGs was sliding back for the first time in many years due to the impact of COVID-19. AI and other emerging technologies were described as a way of catching up and mitigating the negative impact of COVID-19 on global development. AI was also described as having a life-saving and a life-enhancing capacity, and as a key in developing health solutions in response to COVID-19. Using chatbots for frontline responses to COVID-19 queries (a concrete example of AI application) was mentioned on at least two occasions during the launch events.  

Second, values that should guide the development and application of AI were touched upon in almost all interventions. Mr Max Tegmark (President, Future of Life Institute) expressed a key concern, shared by many, when he said that AI might develop faster than the wisdom with which we manage it. There was a sense that our values should shape technology and not vice versa. Trustworthiness, and associated with that, transparency, reliability, and the accountability of AI, was a thread that bound many of the interventions together. Yet, Mr Mattia Fantinati (Member of Parliament and Special Advisor to the Minister of Technological Innovation and Digitalization, Italy) suggested that trustworthiness in itself is not enough, and that people need to see clear benefits from AI application in their personal and professional lives. Similarly, human-centric approaches to AI echoed in many of the interventions. The respect for and protection of human rights were reflected on in many interventions, in particular from representatives of European states and by Director General of the UN Educational, Scientific and Cultural Organization (UNESCO) Ms Audrey Azoulay. She also suggested that sustainability, dignity, and solidarity are under-represented ethical dimensions in the discussion.

Third, in terms of working towards these values, many of the interventions stressed inclusiveness. Ranging from reminders to include small and developing states, to calls for capacity development, as suggested by Ms Irene Solaiman (Policy Researcher, OpenAI), would enable active participation in discussions on AI. Defending and respecting democratic principles when shaping AI values and facilitating a better dialogue on AI, as suggested by Mr Cédric O (Minister for Digital Affairs, France) and MS Rebecca Finlay (Vice President, Canadian Institute for Advanced Research), were echoed by a number of speakers.

Fourth, some suggested that AI values should be universal, while others, such as Mr S Iswaran (Minister for Communications and Information, Singapore) cautioned that values differ. Rules that are too descriptive should be avoided, and applications of AI should adapt to societal norms. Mr Yi Zeng (Director, Research Centre for AI Ethics and Safety, Beijing Academy of Artificial Intelligence) suggested that there are differences, but also common ground, between the philosophies of the East and the West, and that joined efforts are needed to bridge the differences.

Fifth, in terms of AI governance and the relation between global and national levels, there was a sense that complex issues, such as AI governance, require global co-operation. Hochschild suggested that, in contrast to developments on national levels, organisations on the international level, including the UN, are still playing catch up. However, he reiterated that for national efforts to be truly resilient, they need to be complemented by international norm-setting, and added that this is where we have a deficit at the moment.

Some initial reactions and comments

For Chatham House, the Roadmap represents a ‘welcome change in pace and ambition’. The organisation highlights a number of positive developments associated with the Roadmap, such as:

• developing metrics and providing data to monitor and chart progress towards, for example, meaningful digital connectivity and digital inclusion;
• creating a ‘common language’ among stakeholders;
• promoting digital public goods; and
• highlighting the dangers of technology-driven surveillance.

Chatham House’s reaction also points to one of the most profound challenges awaiting digital governance: ‘the reality of profoundly differing approaches between democratic and authoritarian governments’.

Access Now was the co-champion of Recommendations 3a and 3b on digital human rights. In their response to the Roadmap launch, they emphasised three key priorities:

• increasing representation and participation of women and Global South organisations;
• strengthening, and avoiding duplication of, existing human rights mechanisms; and
• highlighting thematic areas, such as Internet shutdowns, digital ID, and cybersecurity that require urgent attention via a human rights-based approach.

Overall, power, class, politics, and race, as drivers of inequality, need to be addressed if digital co-operation and the full enjoyment of human rights and the SDGs are to become a reality. Regarding the tech envoy, Access Now emphasised that much depends on the quality of the person appointed and that ‘civil society must be included in the selection and the execution of the Tech Envoy role’.

Commentators, such as Mr Wolfgang Kleinwächter, issued words of caution regarding the potential of digital collaboration, arguing that the secretary general’s call for a global ban on LAWS ‘is falling on deaf ears’ and that ‘reality also includes the recognition that we probably have to face a decade of bitter conflicts in cyberspace’.

DiploFoundation’s Mr Jovan Kurbalija suggested that the Roadmap is the realistic acceleration of digital co-operation. According to Diplo’s analysis, the Roadmap as a whole puts its main focus on development issues. In terms of AI, the Roadmap clearly recognises governance gaps and suggests a cautious approach by ‘taking into account existing mandates and institutions’. Kurbalija also argued that the Roadmap represents an important shift away from treating AI as a ‘separate policy area’ by ‘anchoring it in existing rules, covering issues such as human rights, liabilities, data protection and consumer protection’. (For further resources and summaries, please visit our dedicated page on the Geneva Internet Platform’s (GIP) Digital Watch Observatory.)

What’s next?

A lot depends on how some of the entities suggested in the Roadmap, such as the tech envoy and the multi-stakeholder advisory body on global AI co-operation, will become operational, and what resources, support, and broad acceptance they will receive. It is also clear that the regulation of AI at the global level remains elusive, given the differences in societal norms and what societies and countries deem acceptable. A ban on LAWS, following the call from the UN secretary general, is out of reach. Yet, developments such as UNESCO’s ongoing consultations on the ethical dimension of AI are hopeful signs. Similarly, suggestions brought into the discussion at the Roadmap launch for a global compendium of AI-use cases and an ‘AI for SDG’ think tank (for the development of free and globally available resources) should be recognised as steps towards more collaborations and the responsible use of AI.

Transportation as the growth engine of the economy has been one of the industries that has been hit the hardest by the COVID-19 pandemic. The pandemic has substantially impacted how we work, how we travel, and how we use technology. It has put an incredible strain on global supply chains, from medical supplies to household goods, as spikes in demand stress-test logistics infrastructures. 

We are now at a crossroads which represents a good opportunity to rethink our modes of transportation. Autonomous vehicles are already used to alleviate the strain on existing delivery services while addressing the demand and reducing the risk of exposure for citizens.

Sustainability in transportation starts with autonomous vehicles; this pandemic has been a game changer for autonomous vehicles in every aspect and has highlighted the significance of the deployment of autonomous vehicles further. 

For instance, the Mayo Clinic has teamed up with the Jacksonville Transportation Authority and self-driving start-ups Beep and Navya for a project in Florida. In this project, autonomous shuttles began servicing a route between a drive-through testing site and a processing laboratory at the Mayo Clinic’s Florida campus on 30 March. Basically, autonomous vehicles are moving COVID-19 tests from a drive-through testing check-point, to a lab for analysis, all without a human on board.

Moreover, Starship Technologies has deployed a fleet of 20 autonomous on-demand vehicles in Fairfax, Virginia, and these vehicles will deliver food and groceries from a handful of restaurants and markets in and around the city’s downtown area. These vehicles have separate insulated areas for hot and cold items and are equipped with cameras, sensors, and other technology to help them glean traffic patterns, curb-cuts, and other information about the urban environment they find themselves in.

In California, the Department of Motor vehicles recently authorised Nuro R2 to test driverless delivery vehicles in some parts of the Bay Area. Nuro’s autonomous vehicle was originally designed for outdoor package delivery. However, the R2 is now supporting two medical facilities, one at the San Mateo Event Center, and the other at the Sleep Train Arena in Sacramento. It delivers linens and medical supplies, as it moves down aisles filled with patient cots during the pandemic.

Another autonomous vehicle start-up Kiwibot delivers safety and sanitary products to students in Berkeley and Denver. Cruise is now using its autonomous vehicles to deliver meals to local recipients and they have made more than 1200 contactless deliveries to low-income, senior citizens from the food bank, and 2500 meals from local restaurants to several organisations serving in the San Francisco Bay Area.

Although the use of autonomous vehicles for human transportation are still uncertain and controversial, as well as very lengthy in terms of the regulatory process, the use cases mentioned have proven the worth of autonomous vehicles for deliveries. It is clear that the technology is well advanced and very useful. Autonomous delivery vehicles have the potential to become part of our everyday lives in a post COVID-19 world.

About the author

Sebnem Tugce Pala is currently working in the hi-tech start-up AmpUp’s public policy team in San Francisco. She is also a researcher at the Transportation Sustainability Research Center (TSRC) at UC Berkeley. Her research interests include micromobility, microtransit, shared mobility, autonomous vehicles, unmanned aircraft systems, urban air mobility, and transportation electrification, and she hopes to apply her public policy experience to the field of new and innovative mobility.

She can be reached at: https://www.linkedin.com/in/sebnemtugcepala/

Artificial intelligence (AI) is a broad term that encompasses high-end technologies capable of ‘performing human-like cognitive processes such as learning, understanding, reasoning and interacting’, according to Szczepanski. Nowadays, societies are exposed to AI through smartphones, virtual assistants, surveillance cameras able of recognising individuals, personalised advertising, and automated cars, to name but a few examples.

In short: 

• The economic potential of AI is huge, but so are the risks that come with it.
• Recent developments on ethical, human rights compliant, and explainable AI are important for keeping technology in harmony with society’s values.
• A flexible but effective regulatory framework is needed to keep AI under scrutiny.

According to McKinsey Global Institute’s 2018 report, AI has the potential to inject an additional $US13 trillion to the world economy by 2030. In general terms, this immense amount of wealth is in itself a desirable outcome of technological progress. Nonetheless, the report also warns that AI, under the current circumstances, is likely to deepen the gaps between countries, companies, and workers and that this ‘needs to be managed if the potential impact of AI on the world economy is to be captured in a sustainable way’.

As any other instrument in human hands, AI can be simultaneously a source of positive outcomes and bring about trouble. The originality of AI, however, the thing that sets it apart from previous innovations, is its pervasiveness and the fact that in order to function it needs to process big amounts of data, much of it inadvertently collected from people going about their daily lives. This in particular needs to be regulated to make sure that AI developers respect the privacy of this data. People have the right to know what information is being collected from them and for what purpose.

Moreover, AI, especially through machine learning, creates devices capable of making autonomous decisions that can have an impact on people’s lives. For example, an autonomous car is able to decide whom it would rather hit (and probably kill) in the case of an accident, where the circumstances leave it no other option, according to a set of instructions embedded in its system.

The ethical dimension of AI has become more visible after public scandals showed to what extent systems could harm people with their decisions and their bias. The Cambridge Analytica case, and the bias detected in recruitment algorithms (that would discriminate against applicants from minority groups) or banking services (that would reject credit applications from members of certain groups) are iconic in this respect.

The Moral Machine, an experiment run by MIT, shows in what types of situation AI would need to make a decision with ethical or moral implications. By using these kinds of models, scientists highlight that much of the time there is no easy answer to an ethical dilemma. When an automated car has to decide between hitting a pedestrian crossing the street, or hitting a wall and harming its passengers, what is at stake is the machine’s ability to discern the lesser evil. But identifying the lesser evil is a tricky business since there aren’t any universal definitions to guide moral reasoning. Data collected by the Moral Machine shows that people from different cultural backgrounds:

have a preference for different values when they have to decide what the most acceptable outcome of an unavoidable tragic situation, like a car accident, would be. The complexity of teaching machines how to reason ethically lies in the difficulty to predict a comprehensive catalogue of situations raising ethical dilemmas, and to agree on what the better course of events for each of those situations is. The Embedded EthiCS initiative at Harvard explores this issue and advocates for the incorporation of ethical discussions in the training of scientists, so that they think through these questions while conceiving and developing AI systems. Developers have to be confronted with the need to ethically justify the specific instructions that they put in their systems, and the outcomes they expect to produce.

Another way of making sure AI systems are not harming the core values of societies is to incorporate the respect of human rights in their conception. Unlike the haziness of ethical questions, compliance with human rights standards is facilitated by the fact that these rights are codified and universal.

International organisations, like the Council of Europe (CoE), have promoted a discussion on what the possible conflicts between AI development and human rights are. CoE has come to the conclusion that states play a fundamental role in making sure that tech companies do not inflict damage to human rights and the fundamental freedoms of people. A ten-step guide published by CoE starts with the need to conduct a human rights impact assessment on AI systems.

Technology is a social product, and as such, it integrates values that orientate the way it operates, even if there was no intentionality in the heads of its creators in the first place. A system that violates privacy rules, that causes indiscriminate damage in a person’s life or property in the case of an accident, or that discriminates against certain groups of people should be held accountable, even if there was no direct human intervention at the moment the damage was produced. AI is challenging in this respect, because the notions of responsibility of the analogue reality aren’t always applicable in the digital world.

Societies should not be forgetful of the fact that technology is a product of the human mind and that the most intelligent machines limit themselves to follow the instructions embedded in them by their human creators. There’s human responsibility behind every step of the creation and operation of AI systems. To trust AI, it’s necessary to hold those in charge of the development and use of high-end technologies accountable for the effects that their actions might have on the lives of individuals. To make this possible, it’s important to establish an appropriate governance that includes smart regulation, and independent, transparent, and accountable institutions in charge of enforcing them.

Since AI is a broad, developing field, the institutional framework set up to regulate it should be flexible and capable of evolving over time as well.

The most urgent task in order to make AI trust-worthy should be making this technology understandable to the public. This could be done by imposing on developers an obligation to explain and justify in clear terms, the decisions behind the systems they develop. DARPA’s project on explainable AI (XAI) is moving in this direction. By advocating for AI systems’ explainability, it moves forward to a scenario where more people understand the functioning of AI and are able to keep it under scrutiny. A more accountable AI is, after all, the best insurance against any abuse of this technology that could harm the values defended by contemporary societies.

Jesús Cisneros is a Mexican diplomat, currently in charge of political and multilateral affairs at the Embassy of Mexico in France. He’s a graduate of the National School of Administration (ENA) in France, and holds a Master’s Degree in public communication from the Sorbonne University. He has successfully completed Diplo’s online course on Artificial Intelligence: Technology, Governance, and Policy Frameworks.

November 2019, when the film Bladerunner is set, came and went. How well did this film – made in 1982 – predict the future? Well, we don’t have androids (replicants) that think and look like us, and there are no flying cars (yet). But, technology has progressed greatly, what seemed like science fiction then – video calling for example – is now part of our everyday lives. There are more and more ‘smart homes’ with virtual assistants which can identify our voices, turn on our lights, adjust the room temperature and answer all sorts of queries. Autonomous vehicles are being tested in many major cities, particularly in the United States; one report predicts that ‘Once technological and regulatory issues have been resolved, up to 15 percent of new cars sold in 2030 could be fully autonomous‘. AI is becoming a big part of our everyday reality and we need to look at how it can help humanity and not harm it.

Bladerunner poster. Visualisation available here.

Before I go any further, I have to mention that there is no ‘one’ definition of AI. But, in this blog we’ll use the English Oxford Living Dictionary’s definition which states that AI is: ‘The theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.’  AI today is driven by huge amounts of data and most of it uses deep learning (a type of machine learning) and natural language processing (a branch of AI that helps computers understand, decode, and employ language). Computers are fed large amounts of data to process and are trained to perform human-like tasks, such as to identify speech and images and make predictions.

There are two main types of AI: generalised and specialised. In generalised AI, the computer system would be able to display the same kind of intelligence that human beings do – to combine all the different data it has been fed, learn from it, and make intelligent decisions based on what it knows. This type of AI has not been created yet, and some argue, never will be – as it is impossible to replicate the human mind. Specialised AI – the type of AI used today – is developed to address only a specific goal, for example, to play chess; to analyse medical data; or to drive a car.

How can (specialised) AI help mankind?

AI is already being used in several sectors such as medicine, finance, and could possibly help us reach the sustainable development goals (SDGs).

•      In healthcare, AI is being used in telemedicine/telehealth: A patient living in a remote area who might otherwise not have access to a doctor is monitored (blood pressure, heart, blood sugar) via a wearable AI device, which, if it detects any alarming changes can send the information to a physician. AI is also used in predictive analytics and precision medicine: it analyses a patient’s medical data, genetic history, lifestyle, environment etc. AI can analyse large sets of data quickly and come up with personalised treatment plans for each patient and even predict medical outcomes. An article in Healthcare Business & Technology argues that ‘In medicine, predictions can range from responses to medications, to hospital readmission rates. Examples include predicting infections from methods of suturing, determining the likelihood of disease, helping a physician with a diagnosis, and even calculating future wellness.’ One such example is: IBM Watson for Oncology: a cognitive computing system which uses AI algorithms to generate treatment recommendations for cancer patients.

Designed by studiogstock / Freepik

•      AI is being used by banks for credit assessment and AI-powered algorithms are being used for trading at the stock market. In e-commerce, AI analyses our data in order to find out our preferences and propose similar items we may want to purchase. For example, Morgan points out that it ‘plays a huge role in Amazon’s recommendation engine … Using data from individual customer preferences and purchases, browsing history and items that are related and regularly bought together, Amazon can create a personalized list of products that customers actually want to buy’.

•      ‘AI for social good’ means that AI could be used to help society, such as by helping us reach the SDGs. According to one discussion paper, ‘Through an analysis of about 160 AI social impact use cases, we have identified and characterized ten domains where adding AI to the solution mix could have large-scale social impact. These range across all 17 of the United Nations Sustainable Development Goals and could potentially help hundreds of millions of people worldwide. Real-life examples show AI already being applied to some degree in about one-third of these use cases, ranging from helping blind people navigate their surroundings to aiding disaster relief efforts.’ 

Some challenges

AI is not perfect and is often biased. While writing her thesis at MIT, Joy Buolamwini discovered gender and skin-type bias in commercial AI systems. These biased systems are being used not only in ‘harmless’ everyday apps, but also in AI face recognition software for surveillance. Some countries have video surveillance on their streets that uses face recognition software. The argument is that it reduces crime. But what if the software makes a mistake? And what if it is used to target people by race? Or to target people who attended a political protest? What about our right to privacy? Our images are being stored without our consent.

There are also privacy and ethical issues to consider when it comes to the gathering, storing, and sharing of healthcare data.

AI can also be used for criminal activities, such as in cyber-attacks and stealing information through phishing – which could lead to bringing down infrastructure such as power stations, hospitals etc.

Another issue is that AI and automation will lead to job loss. How can this be dealt with?

Some solutions

One way to avoid the misuse of AI face recognition software is to follow San Francisco’s lead and ban face recognition surveillance. Some would argue that that is an extreme reaction, and that with proper regulation, such technology should be used. Personally, however, I’m not sure that even with regulations in place, there won’t be significant abuse of such technology.

In order to avoid AI bias such as that discovered by Buolamwini, the data sets used in training an AI system need to be a lot more inclusive, and there needs to be awareness about such biases in order to fix them. In fact, after Buolamwini published her findings, the companies that she had evaluated had quickly made significant improvements in their facial recognition software.

Gender Shades. Visualisation available here.

When using personal data in healthcare, one must tread carefully. Patients must give their consent, and strict regulations need to be put in place to protect the patients and make sure their data is safe and can’t be used against them, by, for example, insurance companies to raise their premiums or deny them care.

When it comes to cybersecurity, it is important to have national strategic frameworks on cybersecurity and AI in place, and there should be regional and international co-operation between national computer emergency response teams (CERTs) and similar institutions.

How to solve job loss due to AI and automation? One of the solutions is upskilling, but not everyone can afford to go back to school. This is why some propose universal basic income (UBI), which is not without its challenges; as Elon Musk pointed out at the World Government Summit in Dubai in 2017. He asked, ‘How will people then have meaning? A lot of people derive meaning from their employment. If you’re not needed, what is the meaning?’            

It is not only low skilled labour that is at risk of being replaced by robots, ‘Jobs at all levels in society presently undertaken by humans are at risk of being reassigned to robots or AI’ according to Bowcott. This is why governments need to make sure that education systems adapt quickly to ensure that future generations are skilled for the kind of jobs that will be available to them alongside AI.

Moreover, new employment and labour legislation are needed. One proposal is for governments to introduce ‘quotas of human workers’. Bowcott suggested that governments could determine which jobs could be performed only by humans, such as, for example, childcare.

Norms and regulations need to be put into place to make sure that humanity benefits from AI and isn’t harmed by it. An international legally-binding treaty on AI is one possibility. The digital world, which AI is a part of, goes beyond borders, and such a treaty would be in the interest of all countries, to protect their citizens. It should take into consideration existing norms and regulations and should be an inclusive process, involve different stakeholders and experts and take their opinions into account in order to come up with a treaty that works for everyone. Lastly, any legislation or treaty related to AI should be constantly evolving and adapting to accommodate developments in the field and avoid ‘law-lag’.

Jelena Dinčić holds a BA in Photography from the Ecole de Condé, Lyon, and a degree in film directing from the Ecole de Cinéma, Geneva. In addition to numerous photography and film projects, she has been working at Diplo since 2015 tackling course coordination, editing, and communications. She has successfully completed Diplo’s online course on Artificial Intelligence: Technology, Governance, and Policy Frameworks.