Thirty years on from IBM’s Deep Blue match against Garry Kasparov, the history of chess still offers relevant lessons about how humans should work with AI. Machines can analyse certain tasks faster and more effectively than people, but that does not render human knowledge and judgment redundant; instead, it elevates different skills. Based on insights from the Polgár Judit Világsakkfesztivál and Morgan Stanley’s experience with generative and agentic AI, here are five guiding principles for making decisions today.
Background: Deep Blue and the continuing question
IBM’s Deep Blue was a chess-specific system that lost to Garry Kasparov in 1996 but won the rematch in 1997. Although its architecture differed from today’s generative AI, the core dilemma remains: if a machine analyses positions more effectively, which tasks should we delegate to it, and where is human judgment still indispensable?
The financial sector: deployment and accountability
Morgan Stanley was among the early financial firms to introduce generative AI tools in wealth management. By June 2026, 98 percent of its employees had access to at least one such tool. The company’s Budapest technology and analytics centre — celebrating its 20th anniversary in 2026 — employs nearly three thousand people and contributes to the firm’s broader technology and AI initiatives. As these tools become widespread, how humans interpret and verify machine outputs becomes crucial.
Morgan Stanley has been the main sponsor of the Polgár Judit Világsakkfesztivál for eight years; the 2026 festival took place on 26 September at the Hungarian National Gallery and focused on critical thinking.
Five lessons for critical thinking
- Receiving a good answer is not enough — you must understand it
Following a machine’s recommended move is different from understanding why that move is appropriate. Players need to grasp the strategy’s rationale, anticipate the opponent’s likely responses, and recognise conditions in which the recommendation could fail. Mayer Dániel, head of Morgan Stanley’s Budapest office, stresses that a faster answer does not automatically yield a better decision: we must know what the answer is based on and what it may have omitted. AI helps process information, but humans must judge whether the data and its quality suffice for a decision.
- Find the weak points in your own ideas
On the chessboard, it is useful to test your own plans by asking what could go wrong. The same practice should apply to AI outputs: which assumptions are least supported, what new information could change the conclusion, and what counterarguments are plausible? You can ask AI for alternative perspectives, but those alternatives must be verified just like the original suggestion.
- A good decision starts with the right question
In chess the objective (win, draw, or maintain position) determines the strategy. In business and daily life, we must also define priorities up front: is speed, lower cost, or long-term trust the most important? If the question is vague, even a confident AI answer can mislead. Mayer recommends framing queries so that responses advance understanding — for example by asking about underlying assumptions and possible alternatives.
- You can delegate tasks, not responsibility
Machines can assist with analysing positions or executing multi-step processes, but final strategic decisions and accountability remain human responsibilities. For agentic AI systems, organisations must define what actions the system may perform autonomously, where errors might occur, and when human intervention is required. Morgan Stanley applies such systems within well-defined workflows, combined with human judgment, oversight, limited access, traceable actions, and rules that protect client data and security.
- The more a machine can do, the more important your own expertise becomes
A player learns most when they not only spot a mistake but also understand why the plan failed and can recognise similar dangers independently next time. The same applies at work: code or analysis generated by AI can be meaningfully reviewed only if the user understands the underlying logic. Technology does not replace domain knowledge; it provides tools that make that knowledge more valuable. Accordingly, Morgan Stanley pairs new tools with employee training.
The human task begins after the machine’s answer
The common thread of these five lessons is critical thinking: understanding answers, scrutinising assumptions, asking precise questions, and accepting responsibility for decisions. These skills were the centrepiece of the Polgár Judit Világsakkfesztivál on 26 September 2026. Morgan Stanley, celebrating its Budapest centre’s 20th anniversary, continues to play a role in the company’s technology development and AI adoption.
Garry Kasparov’s defeat is part of chess history, but the lessons remain timely. Machine performance does not eliminate human thinking; it reframes it. We must understand the recommendations, test assumptions, define goals clearly, and take responsibility for our choices. As Polgár Judit noted, the arrival of AI increases the value of childhood skills that form the basis of critical thinking: play mixes emotions, rules, skill and weighty decisions. Information handed to us is raw material — what we do with it is up to us.
Brief summary
Three decades after Deep Blue, the match’s lesson endures: AI can produce answers, but human judgment must interpret and validate them. The five practical principles outlined here — understand outputs, challenge assumptions, ask the right questions, retain responsibility, and maintain expertise — guide responsible use of AI across chess, finance and beyond.
