#256 Bounded Rationality of Humans etc.

The 3 Big AI Misuse Areas, Decentralisation at all Levels, and a Framework for Modelling Human Behaviour

Course Advertisement: Intake for the 8th cohort of Takshashila’s 48-week Post-graduate Programme in Public Policy (PGP) closes on May 31. It’s a fantastic pathway for those who wish to make a career switch to public policy. Check all details on the website.

---

Global Policy Watch: The Trouble With AI

Global issues relevant to India

— RSJ

We are hitting peak AI boom now. Big Tech is competing between itself to build the best AI platforms. Companies are scrambling to show they are leading the way in AI usage. Countries are playing multi-level geopolitical chess to stymie the efforts of others on AI. And in the interstices, the risk of a runaway, unregulated AI boom is beginning to manifest and mutate itself in ways that raise deep ethical questions.  

A couple of months back I had written about the European Union AI Act that sought to regulate the use of artificial intelligence technology. The Act had set out a comprehensive set of rules to govern AI companies and was seen as a start to making AI safer for consumers. This week, the Act reached the EU Council, and the member states gave the final agreement for the Act to be codified into law. The legislation follows a ‘risk based’ approach, which means the higher the risk of causing harm to the society, the stricter the rules. The governance architecture within the Commission to oversee the legislation would include an AI Office to enforce common rules across the EU, a scientific panel of independent experts to support enforcement, an AI Board for the effective application of the Act, and an advisory forum of stakeholders to help with technical advice to the Board and the Commission. To be sure, the Act will take another 12 months to pass through the parliaments of member countries, and there’s a generous 36-month window for Big Tech to transition their operating model to comply with it. So, things aren’t going to change in a hurry, but the wheels will be set in motion, and it will be interesting to see how the US Big Tech companies will respond to this groundbreaking law. Because the pipeline for AI-related investments is at an all-time high. For evidence, look no further than Nvidia, the poster child of the AI boom. The chipmaker published nosebleed-inducing quarterly results this week. Revenues jumped 268 percent and profits were up seven times over the same period last year even as its CEO spoke of the next ‘industrial revolution’ and companies building AI factories using its chips. 

But this revolution is already burdened with less than savoury aspects of the AI phenomenon. And this is where AI regulations have to move fast to contain the damage. There are three big issues around AI abuse that need to be tackled - a) identity theft, b) disinformation, and c) copyright. 

The last couple of weeks have shown us stark examples of how quickly things are moving in these areas. On identity theft, take, for example, the ease with which a deepfake video conference was used to defraud a Hong Kong-based employee of Arup (a UK-based engineering and design firm) using a fake identity. From CNN:

“Hong Kong police said in February that during the elaborate scam the employee, a finance worker, was duped into attending a video call with people he believed were the chief financial officer and other members of staff, but all of whom turned out to be deepfake re-creations.

According to police, the worker had initially suspected he had received a phishing email from the company’s UK office, as it specified the need for a secret transaction to be carried out. However, the worker put aside his doubts after the video call because other people in attendance had looked and sounded just like colleagues he recognized.

He subsequently agreed to send a total of 200 million Hong Kong dollars — about $25.6 million. The amount was sent across 15 transactions, Hong Kong public broadcaster RTHK reported, citing police.”

Scary.

On Gen AI and copyright, a topic that brought Hollywood to a standstill last year with a writers’ strike, a new chapter was opened last week when actor Scarlet Johansson accused OpenAI of using an “eerily” similar voice to hers for the voice assistant ‘Sky’ that came along with its latest iteration, ChatGPT-4o. Sam Altman, the OpenAI CEO, did not help matters much as he tweeted “her’ after the presentation, which was the name of the Spike Jonze film that starred Johansson voicing the role of a super-intelligent AI. As Reuters reports:

“Even before the latest conflict, agents and executives who spoke with Reuters on condition of anonymity have said for weeks they are concerned that OpenAI’s models appear to have been trained on copyrighted works, which the tech company deemed as a fair use because they are publicly available on the internet. That is seen as a major obstacle by some professional directors and filmmakers, who may be reluctant to use a tool built, without consent, on others’ work.”

The broader question is who owns the copyright if AI tools generate almost perfect likeness to the voice or writing style of someone famous. What happens if such content is broadcast widely and monetised without making it unambiguously clear that it is not real? Once copyright protection becomes vulnerable, it is difficult to see how original content will be valued or differentiated.

The third issue, that of disinformation, is the one that has the potential for the most widespread damage in society. The ability of the AI tool to disambiguate the vast spread of information available on the web and not to hallucinate is vital to prevent ‘fake news’ being established as truth for the future. A couple of days back, Google’s experimental “AI Overviews” tool faced social media mockery for its responses. From the BBC:

“Google's new artificial intelligence (AI) search feature is facing criticism for providing erratic, inaccurate answers. Its experimental "AI Overviews" tool has told some users searching for how to make cheese stick to pizza better that they could use "non-toxic glue". The search engine's AI-generated responses have also said geologists recommend humans eat one rock per day. A Google spokesperson told the BBC they were "isolated examples".

Some of the answers appeared to be based on Reddit comments or articles written by satirical site, The Onion.”

Well, I knew that The Onion would one day become a source of truth online.

The logic embedded within the tool isn’t biasing its responses. Rather, the widespread wrong information available online is being picked up as the accurate response. And once this gets accepted, a self-perpetuating cycle kicks off that only makes the wrong information seem to be right because it is quoted all over online. There’s the Indian equivalent here about the origin of the word ‘dosa’, the breakfast staple all over India. Some online humourists made up an origin story involving a cook, rice batter that had gone sour and the term ‘dosha’ (word for ‘flaw’ in Sanskrit and other Indian languages), and soon enough, there were websites carrying it as a true origin story. It is a matter of time before ChatGPT picks it up as a likely origin story, and you will have the reinforcement cycle kick in. The possibilities and the risks of such disinformation this year during the US election cycle are immense and scary. It is worse for India, where digital penetration through smartphones, cheap data and social media addiction is disproportionately higher than levels of education and awareness of the risks of AI. The multifold rise in online financial frauds after the wide adoption of UPI is only a precursor of things to come. There should be a real urgency to think about consumer protection from the downside risks of AI in India.  

It is even difficult to see how effective the EU AI regulations will be on these issues that are mutating into newer forms rapidly. The regulators will find it difficult to match the rate of technology evolution. It is possible to tightly regulate the ‘primary’ tech, which in this case is the underlying algorithm used by AI platforms as envisaged in the EU Act. But the real challenge is how they are going to be used by the millions who build upon these platforms to abuse identity, spread disinformation or infringe on copyright. That genie is already out of the bottle.  

---

India Policy Watch: Centre Shouldn’t Hold

Policy issues relevant to India

— RSJ

One of the things we have often written about here is the problem of over-centralisation in problem-solving and policy-making in India. The latest episode of Amit Varma’s show ‘Everything is Everything’ with Ajay Shah dissects the need for decentralisation in India. As it happened, I watched the show soon after reading the chapter on centralisation from Karthik Muralidharan’s excellent book ‘Accelerating India's Development’, about which we have written in past editions. 

As promised, I will select excerpts from the book on key topics of interest. These excerpts should give you further reasons to order a copy immediately. 

So, here’s Prof Muralidharan on centralisation:

This tension poses a fundamental challenge for India because there are two valid principles that are in conflict. As per the ‘equality of all citizens’ principle, delimitation is long overdue and should be done in full. However, as per the ‘federal’ principle, which treats states as meaningful entities in the national polity in their own right, delimitation would not only be unfair, but also break a promise designed to incentivise states to controls their populations, which the better performing states have done - especially by investing heavily in women’s health and education. 

So, we will need to explore a wide range of solutions and compromises that can adequately balance these and other considerations that are unique to India. In addition to the options of full delimitation or extending the freeze, some intermediate options include: (a) implementing delimitation in the Lok Sabha, but modifying the rules of the Rajya Sabha so that it functions more like the US senate where each state has an equal vote, (b) continuing to freeze political representation based on the 1971 census, but using current census figures for economic transfers, and (c) reforming the Seventh Schedule of the Constitution to substantially increase state-level fiscal and policy autonomy so that a relative reduction in Parliamentary representation has less impact on citizens’ lives in states that experience such a reduction. 

The last point is why the challenge of delimitation can also be an opportunity to revisit and reoptimise India’s federal compact. As shown in this chapter, India is the most over-centralised large country in the world, which reflects unique concerns that the founding framers of our Constitution had. As we enter our seventy-fifth year as a sovereign Republic, much has changed. The political unity of India is far more robust now than in 1959; central planning is much less important for our economic well-being than state-led dynamism; and sharp increase in education and access to information make it easier for local communities to resist elite capture. These changes make it both feasible and optimal to decentralise much more, while maintaining essential checks and balances. 

The greatest cost of over-centralisation may be the weakening of Indian society and democracy by reducing citizens’ agency, engagement, and initiative in solving their own problems at a local level, and instead looking to a distant ‘Sarkar’ to do so. Greater decentralisation will increase the number of people engaged in governance, and reduce the time taken to respond to local problems. The point has been succinctly captured by decentralisation advocate Ashwin Mahesh, who has noted that: ‘The best way to solve our numerous problem is to increase the number of problem-solvers.’

---

A Framework A Week: How to Deter Reasonable People from Engaging in Undesirable Behaviour?

Tools for thinking about public policy

— Pranay Kotasthane

This week, I read a paper that questions the now-common notion that human beings are inherently irrational. This assumption has significant consequences for every domain of public policy, so it’s worth spending some time on it.

The paper, titled Behavioral Science Should Start by Assuming People Are Reasonable, argues that we should start with the premise that humans are reasonable, neither rational nor irrational.

Over the past couple of decades, conventional economics has been under the pump for its base assumption that humans, as self-serving agents, make rational choices, and consequently, their behaviours are governed by the incentives they face. In its place came the behavioural science revolution, which started with the assumption that we often make irrational choices. The ever-expanding list of cognitive biases was supposed to remind us of all the ways in which humans act against their own interests.

But now, there seems to be a growing consensus that it’s better to assume that humans are rational, after all. Not to the extent that their decisions are always informed by a utilitarian cost-benefit calculation but to the extent that most humans try to make optimal use of the available information that fits their own contexts and goals.

Let’s understand this difference with an example that many of us can relate to. Playing the lottery seems irrational at first, given the extremely low probability of winning and the poor expected value of a lottery ticket. Indeed, that’s what my instinctive response was when I observed that shops selling the Singapore Pools Lottery were almost always teeming with people.

When we begin with the assumption that people who play the lottery are irrational, the policy solutions we tend towards are often coercive and paternalistic. Since the starting assumption is that we are prone to making terrible decisions, it’s only natural that an omniscient government should rescue us poor little sods. Some potential solutions that emerge from this perspective are bans on lotteries, mandatory education campaigns to emphasise the low odds of winning, and heavy taxes on lottery winnings.

In contrast, if we start with the premise that people are reasonable, lottery-playing can be explained as a form of cheap entertainment or escapism. For a relatively small cost, a lottery ticket provides the opportunity to dream about a life-changing windfall. Another reason could be the social and cultural context in which lottery playing occurs. In some communities, playing the lottery may be accepted behaviour. Participating in this shared activity could provide a sense of belonging or connection to others. A third reason to play the lottery may be due to systemic factors such as limited upward mobility. For individuals who feel that traditional paths to financial success are out of reach, the lottery may represent a rare opportunity for a transformative change in circumstances. In a society where success is often portrayed as the result of luck or chance rather than hard work or merit, playing the lottery may be seen as a reasonable strategy for achieving one's goals.

The policy solutions, in this case, would then focus not just on individual behaviour change but also on addressing the systemic factors that make this behaviour appealing or necessary for some people. This could include efforts to increase economic opportunity, provide alternative sources of entertainment and social connection, and challenge cultural narratives around success and luck.

Thus, our starting assumption matters. A lot.

The conclusion of the paper is worth highlighting:

“Assuming that people are reasonable highlights the need to engage with people who are the targets of interventions. This turns them from targets into partners and also acknowledges the systems they inhabit. This more participatory and deliberative engagement can manifest in methods such as qualitative interviews, focus groups, citizens assemblies, engagement with citizen science initiatives, and deliberative prompts (‘nudge plus’) during interventions.”

The two fundamental assumptions seem to be related to the “compassion vs empathy” difference I explored a couple of weeks ago. Assuming people are irrational and hence need immediate external help is consistent with an empathetic approach. In our lottery-playing example, this approach could lead to policies that prioritise short-term relief, such as providing financial assistance or increasing access to gambling addiction treatment.

On the other hand, assuming that people are reasonable is consistent with a compassionate approach because this view takes into account the systemic factors influencing a person’s behaviour. This lens could lead to policy solutions such as increasing economic opportunity.

---

HomeWork

Reading and listening recommendations on public policy matters

1. [Article] Stephen Walt, in Foreign Policy, makes a realist’s case for opposing the war in Gaza. Well worth a read.

2. [Article] This article on the role of social norms and deterrence in reducing drunken driving is fantastic. We will have more on this topic in the next edition.

---