The key facts
- Bill Gates told NBC News that AI could become a dangerous weapon in the hands of malicious actors and called for legislative oversight
- Expert David Linthicum considers the catastrophic scenario unlikely: it would require a combination of several factors
- Current risks include data leaks, fraud, deepfakes, and connecting unvetted AI systems to organizations' real systems
Gates's Warning
Microsoft co-founder Bill Gates said in an interview with NBC News that modern artificial intelligence tools could turn into an extremely dangerous weapon in the hands of people with malicious intent. He also stressed the need to control the technology at the legislative level.
Expert's Assessment: a Catastrophe Is Unlikely
To discuss this warning, Tom's Guide turned to David Linthicum, an expert in AI, cloud technologies, and cybersecurity. According to him, for AI to trigger a catastrophe of such a scale, it would need the ability to affect critical infrastructure, weapons, supply chains, or biological systems, and to act without human involvement. For this, the expert believes, several factors must coincide: major errors in system integration, serious governance shortcomings, and deliberate abuse of the technology. Therefore, he does not consider such a scenario inevitable or expected in the near future.
Terrorists and Unstable States
Linthicum suggested that Gates may have meant the possibility of AI being used by terrorists or unstable states. However, in the expert's opinion, there are serious practical obstacles on the way from a malicious idea to a large-scale catastrophe.
Real Threats Today
At the same time, the expert does not deny the current risks associated with AI. He listed data leaks, fraud, deepfakes, improper automation, and biased decisions among the current problems. His main concern is connecting insufficiently developed AI systems to real corporate and government systems: some organizations may implement the technology without the necessary control mechanisms, which can cause harm.
How to Reduce the Risks
To reduce the threats, Linthicum proposed granting AI only the permissions necessary to complete a task, isolating systems, and tying dangerous actions to human confirmation. Also, according to him, it is important to log all system actions, test it regularly, and keep the ability to roll back changes. The expert believes that relying only on an emergency kill switch is not enough: if a person discovers a problem too late, harm may be done before the system is shut down. Therefore, security must rely on multiple layers of control and be considered from the earliest stages of development.
