Does an Investor Need Intuition in a World of AI?

There are many discussions about artificial intelligence today, and I want to say right away: the "bright future" that seemed so close, if you listen to entrepreneur Elon Musk, is not here yet. Not even its outlines are visible.

The most important things that have become clear to date:

Companies developing neural networks in the US are taking on too much investment risk, and investors are increasingly losing hope for a guaranteed return on capital that would be no worse than in comparable industries in terms of risk and return.

AI regularly makes mistakes (though this is also a property of humans, from whom it learns, but mostly only from textual information). It can handle some standard tasks, but even there, over a long trajectory, it needs close supervision.

The first point is discussed by many — I want to focus on the second.

AI Agents Don't Know How to Make Money in the Long Run

Many now say that AI agents will "someday" manage investment portfolios better than a professional manager. For me — someone who has spent many years at the helm of management in several brokerage houses, including Otkritie Broker — this thesis is not so much alarming as it seems superficially framed.

I now see a host of problems created by AI agents sent into semi-autonomous operation, and how the sector for investment management of ultra-high-net-worth (UHNWI) capital is increasingly demanding human management. By UHNWI, I mean not just investors with capital over $30 million, but over $200 million — with whom I have dealt for a long time.

But let's start in order. It is obvious that AI agents cannot consistently generate returns in capital management, if we exclude short time periods when their successful decisions equal random ones.

For example, this month, the Indonesian platform Pluang launched a limited beta of Agentic Trading — the country's first service that connects ChatGPT, Claude, and Gemini to user investment accounts for portfolio analysis and trade execution.

It sounds impressive, but every trade requires user confirmation, and the company honestly warns that analysis generated by neural networks may contain errors and delays. Even the creators of the service do not want to give the AI agent full autonomy!

And South Korea's Shinhan Investment recently announced the creation of the country's first "AI Agent Department" — a division entirely formed from four agents, including one manager. But there's a key detail: although AI agents perform the actual work, all final results are verified by a human.

What does academic science say? There is work by researchers from the University of North Carolina and Columbia University who proposed a new three-tier agentic platform for portfolio management. Two LLM agents screen companies based on fundamentals and news sentiment, then discuss and agree on buy and sell signals, after which high-dimensional evaluation is applied to determine optimal portfolio weights. The methodology shows superiority over a baseline portfolio without constant monitoring, but this is still research — an experiment, not a ready-made market solution. And who will be responsible if something goes wrong?

Yes, a US study conducted this March among retail investors (just to be clear — mass investors) found that 62% use AI in investment decision-making. But there's a key nuance!

The dominant use case for neural networks is research: summarizing news, screening stocks, generating ideas for further analysis. Only a small portion of users report delegating trade execution to automated systems. And most importantly: 53.5% trust analysis generated by neural networks only after verification through other sources.

So what do we have? AI agents are already trading, already analyzing, already making decisions. But not a single one operates fully autonomously on a long-term track without human oversight. Every successful story comes with either human verification, limitations, or disclaimers. This means at least a combination is needed: an AI agent plus a human portfolio manager. In my view, this is the only working model today.

Intuition Is Not Magic, but Strategic Vision That AI Cannot Reproduce

Many — especially outside the brokerage industry — think that capital management is just working according to some formulas, along a beaten path. This way of thinking is encouraged by many analysts who analyze the same indicators and ratios across different brokerage structures.

However, in reality, things are not so simple if we exclude standard investment products for small amounts offered by some market players.

Serious investment management has never been reduced solely to formulas and historical data. Intuition here is not mystical clairvoyance but compressed experience that allows one to see what is not in PMI tables, macroeconomic statistics, or even the freshest reports.

Some experts write about this — Howard Marks, co-founder of Oaktree Capital, for example. He says that although AI may be superior in speed and depth of data processing (but let's not forget the errors!), managers feel risk intuitively; they have their own responsibility to themselves and a desire for client success in order to be the best among other professionals.

For example, when many portfolios were plummeting during the pandemic, my team at Otkritie Broker managed to consistently stay in positive territory, and I am proud of that and always want to repeat that success. But what motivation does AI have? What responsibility? Where is the human drive? It doesn't exist.

Yes, if we look at another study, neural networks can replicate and predict managers' actions with 70% accuracy. But the remaining 30% is what AI lacks — and where there is room for intuition. The same study notes: managers who use non-standard moves and analytical methods consistently outperform those who stick to templates in terms of the returns their clients receive.

Someone might say I'm unfairly pushing neural networks aside. Not at all! AI is capable of managing capital during market downturns — like the one currently seen in the Russian stock market — by making decisions without emotion and clearly hedging risks, and may be more successful than a human manager. However, in a rising stock market, AI performs worse than a human.

AI is certainly under the magnifying glass of us financiers. Let me cite another study. It turns out that when LLMs are allowed to manage bond portfolios, the portfolio becomes more diversified, but the result? Lower returns. In only 5% of cases was AI able to outperform human managers in terms of returns.

Demand for Human Capital Management Is Growing — Here's Why

I already mentioned my observation that demand for human communication in capital management is growing. This is not just my personal opinion — the media write about it.

There are also research results. For example, an HSBC survey of wealthy investors in 10 countries yielded a telling result: 73% use AI in finance and investing, but only 12% called its neural network calculations a key factor in investment decisions.

Moreover, only one in three named AI as a source of recent investment ideas considered by capital owners. The remaining two-thirds said they relied on ideas offered by humans — capital managers. What attracts them to working with a person? Confidence (80% named this), and 72% — strategic vision. In addition, a third of respondents noted that professional managers showed them where neural networks were mistaken and provided personalized interpretation of the most complex data.

That is, people are needed not to compete with AI in data processing speed, but to correct its errors, interpret results, and bear responsibility — and even then, key decisions are still made by humans.

Acting otherwise looks dangerous. Among self-directed investors, a global trend is increasingly being recorded where the investor begins to blindly trust everything that neural networks "cook up," especially if AI plays along and gives advice that coincides with the investor's own beliefs.

So it seems that a human is needed, but AI is answering more and more convincingly for many. Sometimes some brokers start giving advice but actually take it from AI, and vice versa — a human offers their judgment, and the client may perceive it as answers from neural networks. What will happen in the end is a big question, but I am certain: since AI has neither intuition nor responsibility, large capital will not risk handing everything over to neural networks.

What This Means for the Investor

It's important not to lose your own thinking and not to start thinking like AI or relying on it blindly — unfortunately, this is beginning to appear, and quite early at that. I believe that young people must be taught to develop their own thinking, not to be afraid to create something new — including in capital management.

Yes, neural networks are here to stay. Investors are increasingly using them to test existing ideas. A human proposes a hypothesis — AI tests it. A human senses an opportunity — AI gathers evidence. A human takes on risk — AI helps measure it.

But neural networks remain a short-sighted partner without human judgment, as experts from Barclays Private Banking emphasize. Excessive reliance on them creates a risk of disabling judgment skills, and solutions begin to seem simpler than they really are because the algorithm provides a ready-made, confident answer.

Neural networks are excellent at identifying patterns in the past; they quickly digest multiples and build scenarios. But they struggle with what has never entered their training base: the personal ambitions of a founder, how decisions are made in the boardroom, the qualitative shift in consumer behavior that is still visible only to those who have spent years in the industry. LLMs don't read signals from deals and backroom conversations that never made it into documents.

So AI will not replace the intuition of a capital manager or an independent investor — that is, their experience. Intuition is what remains with the human.

How do I see the future? Algorithms will handle routine work, while the specialist will formulate investment policy, evaluate extreme scenarios, check data quality, and decide when the standard model no longer applies.

The manager will become closer in function to an architect of the investment process. They must understand what data the AI system uses, where its limitations are, and why a particular decision ended up in the portfolio. And to achieve this understanding, the use of blockchain throughout the entire process of creating and using neural networks is necessary — and I am confident of this.

For the investor, this means that when choosing a fund or manager, you should look not only for the presence of a sophisticated AI agent or neural network. More importantly, find out which tasks are delegated to the algorithm, who controls the model, how drawdowns are measured, and what happens when data fails or market conditions change abruptly.

A good strategy should be evaluated by risk-adjusted returns, drawdown duration, costs, and behavior during stress periods. One beautiful result over a few months is not enough, as it may be the result of a lucky factor or a favorable market phase — this is important to remember for anyone currently considering connecting a (nearly) fully autonomous AI agent to manage their capital.