Jon Krohn: 00:00
This is Five-Minute Friday on what AI platform companies might be like in a decade.
For the last two Five-Minute Fridays I’ve been joined by Ben Taylor, and we have him again this week answering questions from me on specific important AI questions. Two weeks ago we had him filling us in on how you sell a multimillion dollar AI contract. Last week he was back to let us know why CEOs care about big AI contracts and the big difference that can make for their businesses. Now looking ahead as we did a bit last week into the future and what AI can do for us, let’s dig into that more this week by asking what should we expect from AI platform companies 10 years from now? Could one AI platform eat all of the niche AIs, for example?
Ben Taylor: 01:16
I love this question. With AI platforms, one of the issues you run into is you can start going after every problem, and that is a little concerning because you have limited resources, what’s the customer acquisition cost? Typically people want to do market verticalizations, but eventually there will be a platform that can go after every problem, so any problem that you and I would roll up our sleeves and say, “Principal level data scientist reporting for duty. I’m going to dive in and do this,” in 10 years the platform will do it all, and a few themes that you’ll see, there’ll be more data-type expansions, more data source expansions, so they’ll be able to tackle more problems that previously required your heroics to fill in the gaps, and then it’ll be more approachable by more users.
We talk about democratizing AI. Right now you have to code, but there is a future that I’m very excited about where not only do you not need to code, you don’t even have to know the UX, know the UI. I imagine executives in the future, I call it… We go through this next five to seven years, you’re going through anticipated actions, but then you get into the territory that I call Jarvis for Everyone, so Jarvis for Everyone, yes you can use the UX. Yes you could use the APIs if you choose to.
Jon Krohn: 02:31
Right.
Ben Taylor: 02:31
Or you just talk to it.
Jon Krohn: 02:33
Right.
Ben Taylor: 02:33
So you go into your office, just like you and I might task a group of five or 10 junior data scientists, you are now tasking this intelligent platform because voice is a much more natural way to communicate and we really are there. The accuracy in ASR, the quality of NLP, so 10 years from now you will have these very specific and direct-
Jon Krohn: 02:56
What’s ASR?
Ben Taylor: 02:57
Automatic Speech Recognition. A human… The word error rate substitution inclusion or deletion is, not plus or minus, it’s 3% word error rate. ASR systems in the past have been really bad, we’re talking 20% word error rate, which is hard for you to read-
Jon Krohn: 03:17
Yeah.
Ben Taylor: 03:17
… but that is quite good. With Alexa and these other systems, they’re really good because they control the hardware. If you get below that then it’s… You don’t want bad ASR at the top of the pipeline because that’s going to mess up all the downstream NLP.
Jon Krohn: 03:33
Yeah.
Ben Taylor: 03:33
So when ASR is so good that it can hear everything that you and I are saying-
Jon Krohn: 03:39
Right.
Ben Taylor: 03:39
… with 1% word error rate, then it’s ready to now have a conversation which is… it’s a much more efficient way to interact with any software platform, because otherwise you have to learn the UX, you hope it’s intuitive, and why do a five, six, seven, nine-step process when you can just ask the question, so a fun thought to sneak in there.
Some people don’t like personifying the AI but I think it’s a little charming. You can imagine a scenario where I’m drinking a glass of wine on the balcony, I’m giving some instructions to this AI system, and I’m going to go to bed, but I messed up. I gave it some instructions that were faulty, or something I asked her to do wasn’t quite right, and while I’m sleeping-
Jon Krohn: 04:23
Maybe a little evil.
Ben Taylor: 04:25
Yeah. So while I’m sleeping, this AI system desperately needs to try second, third, fourth, fifth alternative. It needs to proactively unblock itself even though it was my fault. I told it to do this churn model or something, hopefully something much more advanced than that, and so I’ll wake up in the morning and not only-
Jon Krohn: 04:46
Create me some customers who will never churn.
Ben Taylor: 04:50
Yeah. Make my competitors go bankrupt, go to sleep, so that eventually when you have the singularity, it will surprise you. You’re like, “Oh wow, it actually crawled the web and downloaded all this data that maybe was illegal, but I didn’t know it could do that”.
Jon Krohn: 05:04
Right.
Ben Taylor: 05:06
Yeah. So you’ll have to be very specific. I need you to find all data possible without violating GDPR and any local laws.
Jon Krohn: 05:18
All right, so we’ve got this AI system that can not only understand your commands but also understand mistakes that you might have made in your commands.
Ben Taylor: 05:32
Yeah.
Jon Krohn: 05:32
And automatically.
Ben Taylor: 05:33
Yep. And some people might not like this future, but I think it’s a great future because you and I are underqualified. Everyone, everyone that’s listening. You are underqualified for most of the work you’re going to do tomorrow. Wouldn’t you rather spend your time being underqualified where you’re literally learning, where you’re trying these new challenges, and I think that’s a very exciting place to be because what are we going to do? We’re going to do a lot of things that have never been done, but unfortunately today we waste time reading emails, or even some of the machine learning things you might do, you’ve built this bike so many times, you’re just… It’s not building itself yet, and so I’m really excited about that future, and I think there’s plenty of space to play for niche AI players, but I do see some monster platform company that begins to gobble them all up and that would… Not that Mark Cuban… He has this quote that AI will usher in the world’s first trillionaire, so if you have that company-
Jon Krohn: 06:44
Right.
Ben Taylor: 06:45
… that’s running at that level, then it most likely is a $10 trillion market cap company, and so maybe there’s a trillionaire hidden in there somewhere, right?
Jon Krohn: 06:53
Maybe it’s you, listener.
Ben Taylor: 06:55
Yes. Get to work.
Jon Krohn: 06:58
All right. That’s it for today’s episode. We’ll be back with Ben Taylor for the fourth and final time next week to ask him another topic that prepares us for the future, how to get kids excited about STEM subjects.
In the meantime, keep on rocking it out there folks and I’m looking forward to enjoying another round of the SuperDataScience podcast with you very soon.
