In Part 1 of this series, I said something that I’d thought I’d never say: once we’re coping with typical enterprise-scale AI systems, trust and safety is straightforward.
What?! Blasphemy!
Hear me out.
Yes, okay, it’s actually pretty hard. But the issue pales compared to the trust and safety headache that’s the latest wave of generative AI. Here’s why.
Imagine you’re the CEO of an airline without an AI-based ticket pricing system. Your Chief of Staff runs into your office panting that some team of information hotshots in your organization is hours away from a full-scale AI pricing system launch, but they were overheard saying, “I don’t know how good this AI system is. Dunno how much revenue it makes or loses… nevertheless it seems useful, so let’s launch it.”
Heads will roll. Such a system’s reach and potential business impact is just too massive for this level of sloppiness. You’ll likely fire everyone who had anything to do with this completely unhinged scenario and also you’ll be right to do it. In spite of everything, because the CEO, ultimate responsibility for the airline’s success falls to you and eliminating this gaggle of clowns can be a no brainer given the inappropriate level of risk they almost subjected your enterprise to. The entire situation is criminally silly. Your organization is healthier off without them.
Say what you’ll about large organizations, however the one thing they have an inclination to be good at is avoiding anything that frivolously rocks the boat.
Typically, an issue like that’s smothered long before it reaches the CEO’s desk. Say what you’ll about large organizations, however the one thing they have an inclination to be good at is avoiding anything that frivolously rocks the boat. There’s a built-in preference for caution over gambling, which is why an enterprise-scale AI system typically only gets out of the gate if (1) it provably solves a problem provably well or (2) it has a provably low potential for harm (since the stakes are low, because errors wouldn’t be very embarrassing/painful, or because the applying is of low strategic importance).
The straightforwardness of the AI system’s raison d’être is a particularly powerful simplification tool.
Examples from an airline’s standpoint:
(1) An AI pricing system that’s fastidiously launched in a gradual ramp-up and statistically tested to have a positive revenue impact of not less than x%.
(2) An AI pronunciation system that enables a gate agent hearken to a data-driven best guess about the right way to announce a passenger’s name to assist the agent out in the event that they’re unsure concerning the pronunciation. A system like that is hardly mission-critical and it comes with the upside of with the ability to tell the world you do AI without taking over much risk. Also, harmlessness is simpler to attain when trained humans get to approve all of the output, so that you’d want your gate agents to make use of their judgment.
The purpose is that unless the trust and questions of safety are already minimized by the very nature of the applying, an enterprise-scale AI system isn’t going to see the sunshine of day unless there’s proof that its upside is well worth the risk… and getting this sort of proof is inconceivable by definition when there’s no clarity concerning the value that the system provides.
Why does this make things easy? Since it signifies that every mission-critical traditional enterprise-scale AI system (category (1)) tends to have:
- a comparatively straightforward use case statement
- a vision of what the intended “good behavior” for the system looks like
- a transparent, monolithic objective
- measurable performance
- well-defined testing criteria
- relative clarity about what could go flawed and thus which safety nets are needed
There are many challenges here too, like the right way to guarantee that a system like this plays nice with all the present enterprise systems (see my YouTube course for that and more), however the straightforwardness of the system’s raison d’être is a particularly powerful simplification tool.
The important thing insight here is that the economics for enterprise-grade solutions are likely to favor scale. Systems intended for deployment at scale normally have a transparent purpose, else a sensible leader sends them straight to the rubbish compactor. That’s why most enterprise-grade AI systems of the past decade were designed to do one very specific thing rather well at scale.
Most enterprise-grade AI systems of the past decade were designed to do one very specific thing rather well at scale.
This can be a huge advantage for trust and safety. Huge! Sure, there’s loads of general reliability work to do to be certain that you retain your users secure when your system meets “the long tail” (the bizarre users), nevertheless it’s still quite a bit easier to guard a varied group of users from a single-purpose, single-function system than to guard the identical group from a multi-purpose, multi-function system. And from most enterprises’ perspective, Generative AI systems are fundamentally multi-purpose and multi-functional.
That’s the important thing insight, so let’s repeat it:
It’s quite a bit easier to guard a varied group of users from a single-purpose system than to guard the identical group from a multi-purpose system.
For those who’d like a greater understanding of this insight, proceed on to Part 3 of this series.
Then again, if this last insight is clear to you, then be at liberty to skip Part 3 and head straight to Part 4 where I explain why generative AI doesn’t include these same simplifying characteristics and what which means for AI regulation.
For those who had a good time here and also you’re on the lookout for a complete applied AI course designed to be fun for beginners and experts alike, here’s the one I made to your amusement:
P.S. Have you ever ever tried hitting the clap button here on Medium greater than once to see what happens? ❤️
Let’s be friends! You could find me on Twitter, YouTube, Substack, and LinkedIn. Fascinated by having me speak at your event? Use this kind to get in contact.
jazz study