S11E19 – Where to Draw the Line on AI Decisions

Episode Summary

Mike opened the show by asking where the line should be on letting AI make decisions, from military drones to surgery to who gets fired. He and Howard agreed that a person has to stay mentally involved and be able to push back, because AI tends to take the cheapest path and can produce something unsafe or unfixable.

The rest of the episode centered on security. Mike argued that the weak point is usually the person, not the antivirus, and gave examples including Clover Health, love scams aimed at lonely older people, and a Florida developer who put a game on Steam that stole cryptocurrency. He also covered public charging ports, cheap cables, found flash drives and credit card data kept on paper.

Mike closed by explaining why he started the show and the company, and reminding listeners that the call-in help is free. There were no callers this week.

Topics Covered

Where to draw the line on AI

Mike asked listeners where AI should stop making decisions, pointing to military drones, surgery and firing decisions. His concern was the human becoming just the hands for the machine. Howard added that you cannot let AI run on its own, and that if you did not know the goal you could end up clicking yes to everything.

Building with AI safely

Mike said projects built by letting AI run unchecked hit a point where they become a mess that is impossible to fix. He described a prospective client who built his own system for care homes without knowing which AI helped build it, and warned that if you build it yourself you are responsible for any breach. Internally, Mike said they have several AI tools check each other’s work and ask them to be adversarial, to get a second, third and sometimes fourth opinion.

People are the weak link

Mike said built-in Windows antivirus is fine for most home users, since the weakness is the human element. He said Mac and Linux users are not inherently safer, and that social engineering, such as someone asking for a password reset, is how most break-ins happen. He mentioned Clover Health having an issue this week, as an example where nothing was hacked other than the people involved. He also said love scams are nearly impossible to stop with software, and suggested checking in on older relatives and calling your parents.

Trusting marketplaces and public charging

A developer in Florida released a game on Steam that, according to the hosts, stole more than $200,000 in cryptocurrency by attacking wallets. Mike said being in an app store or on Steam does not mean something is safe, and that VirusTotal is a free place to get a second opinion on a file, though it can give false positives. He also warned not to use public charging ports or the cheapest cables, because a cable can hide electronics that act like a keyboard and steal data.

Planning for the worst and card data

Mike said ransomware is unlikely in any given year but can end a business, so verified backups and knowing how much work a restore would take matter. On PCI compliance, he said writing down the three-digit code from the back of customers’ cards, such as in a binder for recurring payments, is the biggest failure and can get a business barred from taking cards. He said most smaller medical offices are not fully HIPAA compliant before ACG starts working with them.

Found flash drives

Mike said a flash drive found outside, especially in a parking lot, is likely planted, and he advised throwing it away rather than plugging it in. Howard said he plugs unknown drives into a laptop that never goes online. Mike also described testing building security by walking in with a company shirt and asking to be pointed at the server.

Calls

There were no listener calls this episode.