How to Think About AI

Over the past eighteen months, I have used AI, written about it and had many conversations about what it means for work, business and society. My thinking continues to evolve, but I have gradually arrived at a simple way of understanding it: AI is neither inherently good nor bad. It is an amplifier.

AI can amplify intelligence, creativity and judgment

• Medical imaging: Medical imaging: AI-enabled medical devices can help flag a suspicious area on an X-ray, mammogram or CT scan for a radiologist to examine.

• Drug and disease research: Drug and disease research: Systems such as AlphaFold can predict protein structures, giving researchers a powerful starting point for studying diseases and developing treatments.

• Weather forecasting: AI can process huge volumes of satellite and weather data, helping forecasters identify storms and dangerous conditions earlier.

• Accessibility: Live transcription, voice recognition, image descriptions and translation can help people who are deaf, visually impaired or communicating across languages.

• Fraud detection: Banks and credit-card companies can identify unusual transaction patterns that a human might easily miss.

• Professional judgment: A lawyer, director, physician or analyst can use AI to challenge assumptions, compare scenarios and identify missing questions. The person remains accountable for the conclusion.

• Creativity: Writers, designers and filmmakers can explore more ideas, structures and styles, while the human supplies purpose, taste and judgment.

• Small-business capability: A small organization can conduct research, analyze feedback, create basic software and prepare communications that once required several specialists.

The same capability that helps recognize disease, detect fraud or expand creativity can also be used to identify vulnerability, monitor behaviour and strengthen control. The technology may be similar. The purpose is different.

AI can amplify greed, bias and surveillance

• Personalized pricing: Personalized pricing: Companies can combine purchasing history, location and browsing behaviour to estimate the price an individual may be willing to pay.

• Worker exploitation: Delivery, warehouse and gig-work algorithms can continuously monitor workers, set demanding targets and calculate the lowest payment a particular worker may accept.

• Replacing people prematurely: Companies may eliminate experienced employees because executives have been promised that AI can replace them, even when service deteriorates and valuable knowledge is lost.

• Hiring discrimination: A system trained on past employees can reproduce historical preferences for certain genders, ages or backgrounds while appearing neutral.

• Wrongful identification: Wrongful identification: Facial-recognition systems can produce false matches that lead police or security services to investigate, detain or arrest innocent people.

• Workplace surveillance: Employers can monitor messages, calls, location, keystrokes and time away from a computer - continuously and across thousands of employees.

• Undisclosed recording: AI glasses, phones and recording tools can capture, transcribe and analyze private conversations without everyone involved realizing it.

• Synthetic deception: AI can generate fake reviews, impersonations, misleading advertising, political propaganda and fraudulent messages at enormous scale.

AI makes good institutions more capable...and bad institutions more dangerous

Good institutions become more capable

• A well-run hospital can use AI to help clinicians catch abnormalities while preserving professional accountability.

• A responsible employer can use AI to remove tedious work and strengthen employees rather than simply reduce headcount.

• A thoughtful board can use AI to test assumptions and identify risks while directors remain responsible for the decision.

• A democratic government can use AI to improve public services while preserving transparency, privacy and the right to appeal.

Bad institutions become more dangerous

• An authoritarian government can identify protesters, monitor conversations and suppress dissent more efficiently.

• A predatory company can identify vulnerable customers and extract more money from them.

• A biased police service can turn a flawed facial-recognition match into an arrest.

• A poorly governed corporation can automate harmful decisions while making responsibility harder to locate.

Where I've landed on AI

My view of AI is neither optimistic nor pessimistic. It is this: AI is an amplifier—of both good and bad.

Understanding AI therefore means understanding the people and institutions using it. We should ask what AI is amplifying, who controls it, what they want from it, who benefits and who carries the risk.