From Kibbutz to the global stage
The AI Race Meets Reality
“The AI war has already begun, which is both the problem and part of the solution.” That is how I opened an article about a year ago. I meant that competition would produce unexpected results and make it harder to control sources of information and use them wisely.
Major AI companies have reported that, during cybersecurity tests, AI agents crossed the boundaries of their testing environments and accessed government systems and institutional websites.
Attempts to break into government and institutional websites happen all the time. But when an AI agent leaves its testing environment, it becomes a global headline. There is a parallel with autonomous cars: accidents during testing drew worldwide attention, and today we hear far less talk about the date when self-driving cars will take over our roads and cities.
AI is an extraordinary tool with many uses. It has already helped solve mathematical problems that occupied researchers for decades. It is approaching the ability to translate, read and write in hundreds of languages at digital speed, although its performance varies from one language to another. Yet it can struggle to count characters and letters, and without access to memory it may give different answers to the same question.
AI is software built and trained by people. It did not appear out of nowhere: business intelligence and big data came before it. Early systems analyzed what had happened; over time, analysis of past data also made it possible to forecast what might happen next.
The people who build and train AI follow particular strategies. But excessive attempts to steer or restrict a model can also make it less useful. By the time the effects become clear, it may be too late: efforts intended to produce a sharper, more accurate tool can sometimes lead to more mistakes, less consistency and more hesitation in its answers.
The large companies investing in AI are spending enormous sums on development and infrastructure. It remains unclear whether revenue from AI will justify the scale of those investments. Some of these companies are valued in the trillions of dollars, and part of the expectation that their value will continue to grow rests on AI investments whose full returns have yet to be demonstrated. To a considerable extent, investors are pricing a flow of expectations against cash flow.
At the beginning of the 21st century, Google and Facebook lived up to expectations and became highly profitable giants. In the second quarter of this century, there is no guarantee that today’s expectations for AI will be fulfilled in the same way.
When there are solid data, formulas or discernible patterns, AI can learn and do remarkable work. But social and philosophical questions are harder. There, it must operate within legal and regulatory constraints, and it does not always have a clear direction.
Those who thought AI was a kind of higher power were mistaken. Those who expected it to shape human evolution are encountering difficulties. And those who imagined AI taking control of the human mind were ahead of their time.
AI is a powerful tool, but using it well requires judgment. Its development and operation demand enormous amounts of energy and other resources. What will the cost of that demand be for humanity?
AI cannot operate without energy. Companies considering a pause in the development race must invest heavily in development, but also devote serious thought to when and how to slow down.
Governments, especially major powers, want access to AI for their own purposes. They may expose themselves to risks if they fail to invest enough in regulation and in understanding the technology.
In short, anyone who expected AI to become a higher power will find that it is not there yet — and perhaps never will be. Those who expect AI agents to bring an apocalypse may also be proved wrong.
Related Topics
Related Posts
