Haroon Sheikh on AI and Ethics: ‘AI as a Legal Instrument Requires Proper Human-Machine Interaction’

Haroon Sheikh, endowed professor of Strategic Governance of Global Technologies at VU University Amsterdam, expects that the development of artificial intelligence (AI) will greatly impact almost all organizations as well as the lawyers working there. “Their work is about to change profoundly, but they will definitely not become redundant,” says the Senior Scientist at the Netherlands Scientific Council for Government Policy (WRR), philosopher, political researcher and columnist of NRC newspaper.

By Joris Spuesens

Ever since humans learned to use fire, technology has had a profound impact on us. Fire transformed people into technical creatures: it allowed them to prepare food they previously could not eat, they got shorter bowels than other mammals, and they began walking upright. Sheikh: “Fire shaped our bodies. According to Greek mythology, Prometheus gave us the fire of the gods to survive. Just as we can use fire in good or bad ways, we can deploy AI in better or worse ways.”

Sheikh is convinced that AI as such will not turn against mankind and can help us survive. “We should not attribute human traits, like hunger for power, to AI. The desire for power is based on our biological instincts, not on our intelligence. It all depends on how we design AI and which rules we install in AI applications. Naturally, AI might be abused by people with ill intentions. Engineers may do a bad job designing it, or the design may not be secure. That is why it is important to involve public parties in the development of AI and to invest more into the advancement of AI developments with and for public services.”

In Sheikh’s opinion, AI innovations cannot do without ethics. “Fortunately, in Europe we have been dealing with the GDPR for years, which encourages concepts like Privacy by Design. Around the world and in Europe, recent AI developments have sparked many public debates. Some topics of debate are intellectual property, what to do with data from open source applications, who is or will become the owner of generative data, where such data will be stored, and of course how the current and future profits to be gained with AI will be divided.”

The rapid development of generative AI, which produces its own content out of existing data, is a new challenge that has already led to a wave of lawsuits on copyrights. According to Sheikh, the virtual impossibility to identify the source data of generative AI bots poses a challenge to traditional copyright law. “France and Australia have already forced tech companies like Facebook, Netflix, YouTube, Spotify, Apple and Google, which share content from traditional media companies, to deposit money into a fund that supports the original content creators. I expect the same will happen for AI content. Slowly but surely, governments will strengthen their hold on this.”

With laws like the GDPR, DMA/DSA and the AI Act, the EU is a leader in regulating rapid technological developments in the AI field. Sheikh thinks that American tech companies like Facebook, Amazon, Microsoft, Google and Uber also understand that their initial growth strategy of ‘move fast and break things’ is no longer acceptable. “Following that strategy has become too dangerous. The fines imposed by the EU on tech companies that break European rules are already in the billions.”

Sheikh doubts whether strict regulation will put a brake on the development of AI in Europe: “Europe has chosen a different path from simply making money out of AI products and services. The EU wants to deploy AI for important public objectives, such as education, the legal system and health care. Universities, like the University of Amsterdam with its Civic AI Lab, are researching socially useful objectives for AI. These objectives may vary by sector and domain.”

He continues: “We do have to take into account that we are still largely dependent on U.S. cloud providers for the AI computing power we need. This market is dominated by Amazon Web Services and Microsoft, with Europe lagging behind in the AI field. In 2023, the five major players – Microsoft, Amazon Web Services, Salesforce Inc., Google and Oracle – accounted for 67 percent of global revenues, as appears from figures of Statista.”

According to Sheikh, Europe cannot compete with the billions that companies like OpenAI and Google invest in LLM (Large Language Models). “We would do better to focus on developing national applications and building expertise on the artificial intelligence of the large providers, in order to make European countries less dependent on those parties in the future. We could build Dutch-language AI language models with a focus on specific sectors, such as the legal sector. After all, the law in the Netherlands and Europe differs from that in the US, China and Russia. The EU's dominant position will make companies from other countries comply with European rules as a matter of course, as appears from the book The Brussels Effect by Anu Bradford from 2020.”

In Sheikh's opinion, complex legal cases are not yet suitable for AI: “AI is good at analysing historical case law, so it is important to design a proper human-machine interaction to use AI as a legal tool. Thanks to AI, standard contracts for which many data with little variation already exist can be drafted much faster. For example, 50-minute jobs could be sped up by a factor 10 to 100, but an expert will always have to check the job for accuracy afterwards.”

Incidentally, the Dutch police and railway company are already investing heavily in their own AI applications, and EU privacy rules compel them to store their relevant data in Europe or the Netherlands. However, the development of a European cloud, such as Gaia-X, is still too slow to catch up with the United States and China.”

Sheikh points out that Europe has other assets in the technology race against Asia and America: “Thanks to ASML’s chip-making machines and Ericsson’s and Nokia’s network technology, Europe still plays an important strategic role. However, a party like KPN cannot invest in cloud technology at the same levels as the Americans or the Chinese. That is why I do think we need to work harder on creating our own European cloud, in order to prevent geopolitical dependence on other continents." Sheikh elaborates these ideas in articles and in his new book Atlas van de Digitale Wereld (Atlas of the Digital World).

In-house lawyers will have to focus primarily on managing the risks posed by AI. Sheikh: “A problem of AI is that it can hallucinate; it can make up data of its own that are close to previous data. This has already prompted one U.S. Court to make it mandatory to mention that AI was used in legal actions, and that lawyers who use AI always remain personally responsible for the data they provide. AI will mostly make tasks redundant rather than jobs. Some tasks will disappear, but AI will also create new tasks, such as the checking of contracts that were drafted using AI. Such AI applications will save time, which time organizations can then use to build new and different customer relations.”

It goes without saying, he says, that lawyers should first study the GDPR, DMA/DSA and the AI Act to prepare for AI developments. Sheikh: “Organizations can then establish clear principles on how to deal with intellectual property, data access and data use, so that employees cannot just enter sensitive data into AI bots. Once entered, such data cannot be removed anymore and will take on a life of its own. General Counsels will have to state clearly to employees what is allowed and what is not. Crash barriers should be placed, but there should also be room for experimenting in order to drive innovation. AI is far too complex for a top-down approach. Ideally, employees should be allowed to decide for themselves how AI can help them work more efficiently. Let organizations build their own decentralized expertise. Just like people, AI depends on shared knowledge, which arises naturally on the shop floor. You should use the learning curve of employees to develop useful AI applications faster."