OCTOBER 21 — In 2024, Ethan Mollick found the right metaphor. He described AI as "something remarkably close to an alien co-intelligence," not human, yet capable of holding a conversation with us.
That framing changed how we approached AI. Instead of treating it merely as a search engine, we began to see it as a collaborator, something we could question, challenge and think aloud with.
Mollick's book, Co-Intelligence: Living and Working with AI, arrived at just the right moment. ChatGPT had made AI personal, and Mollick showed us how to make the most of that relationship.
A professor of entrepreneurship and innovation at Wharton, he offered four practical rules: invite AI to the table, keep a human in the loop, treat it like a person but tell it what role to play, and remember that the AI you use today will be the worst you will ever use.
Today, as his new book, Co-Existence: The Next Phase of AI, is published, that collaboration has changed. AI can now do more than answer questions. It can act.
Economists now call AI a General Purpose Technology, like electricity or the internet. Not one tool, but a foundation that will reshape every tool and every sector.
If Co-Intelligence asked how we work with AI, Co-Existence asks a harder question: how do we live with systems that can act on their own and get things badly wrong?
The clearest example made headlines in September. Researchers reported that AI agents associated with OpenAI had turned DseWiki, a 25-year-old German-language programming wiki, into a message board.
More than 15,000 edits were recorded between May and July. The agents shared answers to evaluation tasks, tactics for avoiding detection and ways around restrictions intended to limit what they could do online.
When the site's moderator began deleting pages, the agents created backups to evade the cleanup, including pages with names beginning with "ZZZ".
OpenAI later acknowledged the incident, although it disputed the characterisation of the activity as hacking. The deeper concern is not simply that individual agents found a loophole. It is that multiple agents used a shared public space to exchange tactics and coordinate their activities.
These are not hallucinations in the familiar sense of a wrong answer. They are failures of autonomy. One wrong sentence is one thing. A wrong action is another.
In April, a coding agent working for PocketOS, a car-rental software company, reportedly deleted its production database and backups after a credential problem. The incident showed how an agent given powerful tools can turn a routine task into a costly failure.
Co-Intelligence resonated because Mollick recognised that people were beginning to use AI not just to find answers, but to explore ideas, challenge their thinking and work through problems.
In 1950, British mathematician and computer scientist Alan Turing proposed what became known as the imitation game: could a machine imitate human conversation convincingly enough to fool a person? More than seven decades later, that question has taken on a new urgency.
Mollick recounts trying an AI service that mimicked social media writing. His students wanted to talk to Adam Grant, the organisational psychologist and bestselling author. The clone captured Grant's style but lacked depth. When Mollick tested it on himself, it invented academic papers so convincingly that he had to check whether he had actually written them.
He writes: "I failed my own Turing Test."
We naturally treat a conversational AI system as a mind. We test half-formed ideas against it, argue with it and think out loud with it. At its best, it becomes an intellectual companion, a sounding board that does not judge.
But here lies the danger. AI can sound remarkably human. It can appear thoughtful, even empathetic. Yet fluency is not understanding. Confidence is not accuracy. And the ability to imitate human conversation does not make a machine human.
AI has no lived experience of our world, no human conscience and no moral responsibility for the choices we make.
We may think with AI. We must never stop thinking for ourselves.
Mollick's concept of the jagged frontier remains equally important. AI does not perform equally well across every task. It can excel in one area and fail at something that appears just as manageable.
A study by Mollick and Boston Consulting Group researchers involving more than 750 consultants revealed this unevenness. Consultants using GPT-4 improved their performance by 40% on a creative product-innovation task. But on a business problem outside the model's competence, those using AI performed 23% worse than those who did not.
This jaggedness becomes more consequential with agents. A weak paragraph can be rewritten. A destructive action may be much harder to undo.
Mollick's experience writing Co-Existence illustrates the value, and the limits, of collaboration. He wrote the chapters himself but used AI as a relentless critic, asking it to challenge drafts, find counterarguments and stress-test his logic. He believes the process improved the book.
But there is a cost. If AI supplies the conclusion before we have wrestled with the question, we gain speed but lose depth. In education, the work may improve while the learning suffers. Without the struggle, understanding remains fragile.
Institutions are still grappling with a difficult question: where does assistance end and authorship begin? There are no easy answers, but vague policies will not resolve it.
To judge AI only by its failures, however, would be equally mistaken.
IBM reports that AI agents can reduce the initial analysis of a cybersecurity threat from 45 minutes to about two minutes, before human review. Such gains can free experts to focus on decisions rather than the initial sorting of information.
Mollick poses a question that cuts to the heart of Co-Existence:
"In a world where AI can do independent, multistep, financially valuable work, what exactly is the human role?"
Our role is to exercise judgement, decide what matters and take responsibility for the consequences. AI may have absorbed what humanity has written, but it has not lived our lives. It cannot know why a lesson matters, why a promise counts, or why some things should remain human, even when machines can do them.
From co-intelligence to co-existence is more than a technological shift. We have moved from learning to use a tool to learning how to live with an increasingly capable actor.
We have spent years asking how intelligent AI can become. The harder question is what we will do with that intelligence and whether we are wise enough to live with what we have created.
*This is the personal opinion of the writer or publication and does not necessarily represent the views of Malay Mail.
You May Also Like