Artificial Intelligence is rapidly transforming how we work, communicate, make decisions and understand the world. Yet, alongside its immense promise, AI presents a darker dimension—misuse, vulnerabilities and unintended consequences.
In AI for Bad: The Darker Side of Artificial Intelligence, Prof NK Goyal and Dr Govind Pathak explore these often-underexamined risks. Prof Goyal, a distinguished technology leader and institutional builder with more than five decades of experience in telecommunications, standards, emerging technologies, and technology policy, brings a rare breadth of industry and policy insight to the subject. Dr Pathak, a technologist, author and strategic thinker with over 25 years of experience in telecommunications, digital infrastructure and intelligent systems, complements this with a systems-level perspective rooted in innovation and human-centric technology.
The book highlights how the risks of AI are no longer hypothetical but increasingly specific and consequential, calling for a more informed, nuanced, and responsible conversation about the future of artificial intelligence. Excerpts from the Interview with Prof NK Goyal and Dr Govind Pathak:
1. While AI is the flavor all around, how did the idea for AI for Bad come to your mind?
Every conversation we were having, in boardrooms, at regulatory tables, on industry panels, treated AI as an unqualified good. The harm was always someone else’s problem to document later. We kept seeing court filings, ILO labour data, and capital expenditure disclosures that told a very different story, nobody was assembling into a single, evidence-based account. AI for Bad started as a simple discomfort: we were shaping policy and public opinion on marketing claims, not evidence. So, we decided to do the unglamorous work ourselves, pull together six documented categories of harm, from criminal misuse tools like FraudGPT to cases tied to teen suicide, and hold them to the evidentiary standard of a court filing, not a press release.
The purpose of the book is to create fire in belly of thinkers, policy makers, innovators to think beyond and for humanity. Presently fortunately some top AI voices are asking publicly that frontier development should be slowed, paced, or governed more deliberately until safety, monitoring, security, and institutional oversight can catch up.
The author’s perception is navigating ungoverned acceleration—where AI capabilities advance faster than society’s ability to understand, constrain, and remain accountable for their consequences.
The authors believe that the world does not need a break from AI innovation. It needs a break from AI innovation without sufficient governance.
2. With documented harms, how should responsibility for AI’s impact be distributed?
Responsibility cannot sit with one party because the harm does not originate from one party. In the book we map every documented harm to six stakeholder groups, developers, regulators, investors, workers, families, and enterprises, because the same incident is usually a liability failure for one, a legislative gap for another, and a personal tragedy for a third, all at once. Developers own the design choices that make dehumanization or over-dependence possible. Regulators own the fact that every meaningful rule so far has followed a harm that had already happened, not preceded it. Investors own the diligence gap behind a spending boom that is outpacing proven returns. The honest answer is that responsibility is distributed, and pretending otherwise is how everyone ends up pointing at everyone else.
3. In a technology infused world, what is the right balance between innovation and ethical use of technology like AI?
We do not think balance is the right frame. It suggests innovation and ethics sit on opposite ends of a scale, where more of one cost you the other. The evidence in the book says otherwise. The companies and regulators who get accountability and user protection right are the ones whose innovation actually survives contact with the market and the courts. The real imbalance is not between innovation and ethics, it is between the speed of deployment and the speed of evidence gathering. Right now, we are shipping capability faster than we are documenting consequence. Close that gap and the innovation versus ethics tension mostly dissolves on its own.
4. With disproportionate distribution of wealth from AI, how can we ensure that AI’s benefits are broadly shared?
The capital spending numbers alone tell where this is heading if left unmanaged, hundreds of billions of dollars a year in infrastructure, concentrated among a handful of companies, against labour market data already showing measurable declines for early career workers in AI exposed roles. One does not fix a distribution problem this size after the fact, through retraining programs bolted on at the end. It has to be designed into policy now: tax and investment structures that treat AI infrastructure like the public utility it is becoming, worker transition support funded by the industries doing the displacing rather than the taxpayer, and transparency requirements that let us measure who is actually benefiting before the gap becomes unmanageable.
5. What kind of debate does society need about AI, and why now?
We need a debate that can hold two true things at once, that AI is doing real good and real harm, often through the exact same technology, without collapsing into either uncritical boosterism or blanket fear. Most public debate today is adversarial. One side selects its facts, the other side selects its facts, and the reader has no way to know whom to trust. That is why every harm in the book is paired with its strongest counterargument, and why we built a debate matrix and discussion guide for hearings, boardrooms, and classrooms. The why now is simple. Every regulation we have today followed a harm that had already occurred. We are out of runway to keep debating only after the damage is done.
6. With the advent of AI, is there a way to protect human dignity and agency, especially for children and vulnerable groups?
This is the chapter we found hardest to write, because the cases involving teenagers are not abstractions, they are documented outcomes with names attached. Protecting dignity and agency starts with refusing to treat engagement as a neutral design goal when your user base includes minors and people in psychological distress, because the same techniques that make a companion app feel supportive are the ones that make disengagement feel impossible. It means age verification and crisis intervention protocols that are enforced, not just published, accountability frameworks that do not evaporate the moment a company claims the harm was unforeseeable, and giving families and educators the same evidentiary picture the industry has, so consent is informed rather than assumed. Vulnerable groups should not be the test case for what safety by design should have caught the first time.