Сan AI be trusted? UCSD scientists on transparency and risks of technology

But not all AI can be considered safe and trustworthy, according to David Danks, professor of philosophy and data scientist, and Lily Weng, head of the Trusted Machine Learning Lab at the Halıcıoğlu Institute at the University of California, San Diego.

"AI is any system that replaces or augments human mental labour. Just as machines have replaced physical labour," Danks explains. - "But I look at AI from a human perspective - as something that helps us think and act in new ways."

Testing - but not trusting

As the professor points out, he is surprised by how willing people are to experiment with AI, especially in the home - whether it's voice assistants or text generators. But it often doesn't get any further than play:

"We're willing to try, but we don't trust. And perhaps rightly so - many AI systems are not fully trustworthy right now."

Lily Weng emphasises: we need to develop not just "smart" AI, but transparent and error-resistant algorithms:

"Responsible AI is one that explains its decisions, is resistant to attacks and is not susceptible to manipulation. We need to understand how it works - otherwise we won't be able to control it."

Invisible AI: from auto to data collection

Danks points out that AI is increasingly operating "behind the scenes". For example, in modern cars:

"When I open ChatGPT, I realise I'm using AI. But in a car, I don't always know how deep AI is embedded - from engine management to possibly passing my data to insurers. And those are already risks."

In his opinion, it is the invisibility of algorithms that requires companies to be fully transparent: "People need to know when and why AI is being used."

Medicine and trust - the challenge of the future

Researchers are particularly hopeful about AI in healthcare. But the risks here, too, are much higher than with chatbots or recommendation systems.

"AI-assisted diagnosis is a powerful tool. But without trust and explainability of decisions, such technologies can do harm," says Weng. - "Our challenge is to make AI not only accurate, but also transparent."

What AI will never replace

According to Danks, there are areas where algorithms are powerless - and will remain so in the near future:

"Any job where emotional interaction is important - whether it's psychology, education, or human care - doesn't lend itself to automation. AI can mimic empathy but is incapable of genuine connection."

He adds that AI does poorly at tasks where it's unclear what counts as success:

"Algorithms are good when the goal is clear. But life is more complex. We often go by feel, we learn from mistakes. That's something AI hasn't learnt yet."

AI is not a storm, it's a construction site

Danks concludes by emphasising: it's a mistake to fear AI as an impending disaster. It is built by humans, and they are the ones responsible for it:

"AI is not a hurricane, it is the future we are building. It's important to remember: technology is not imposed on us from the outside - it reflects our values, choices and mistakes. If we want AI to be honest and useful - we have to make it so."