The new iPhone might portend an underwhelming future for consumer AI

I'm just going to come right out and say it: The excitement over Apple's new AI chip and facial-recog
The new iPhone might portend an underwhelming future for consumer AI
By ARCHITECHT • Issue #137
I’m just going to come right out and say it: The excitement over Apple’s new AI chip and facial-recognition security feature seems overblown. It’s just so … utilitarian.
Don’t get me wrong: Apple’s strategy is very smart from a UX perspective, assuming the Neural Engine gets a full workout. Use deep learning to power tasks like facial recognition (I’m not convinced this is a game-changer), NLP for text messages, image processing and, of course, Siri. In the process, save battery life by minimizing use of the main CPU and GPU, thus making users that much happier.
But at some point, we might expect the company that all but invented the current image of consumer AI—with Siri—to do something else revolutionary. We might be waiting longer than we think. And not just from Apple, but from Google, Samsung and everybody else, too.
The fact of the matter is that while the field of AI—and deep learning, in particular—is advancing like mad, its strong suit is still in what we might call machine perception. Vision, speech, hearing, language, stuff like that. Which is why we now have devices like the Amazon Echo and Google Home, and phones that can understand us, predict our next words and recognize who’s in our photos. Most of these things we could do before; now we can do them easier or better.
However, there’s still a long way to go before these areas are perfected, and it’s not entirely clear where the next big idea in consumer AI will come from or when it will hit the mainstream. If you look at most consumer applications of AI and machine learning today, they’re really step improvements over the status quo. I think a lot of people would say companies are spending an awful lot of time on AI research so we can add funny mustaches to selfies.
That’s not an indictment of anything, but rather, I would argue, a function of how research works. You don’t change the world overnight, and even promising experimental results can take a long time to make it from lab to production. When they do, I have an easier time seeing game-changing applications of AI in fields like manufacturing, logistics, media and health care than in pure consumer spaces. And even there, we’re talking more about process automation and optimization than about, say, anything that would strike an outside observer as revolutionary. At least in the near term.
Adding intelligence into everyday home appliances at scale could actually be a very big deal, but I don’t recall the last time Whirlpool announced a new dishwasher with all the world watching. Who knows: maybe the consumer-AI winter will come not because there’s no meaningful progress, but because consumers grow tired of incremental improvements presented as monumental achievements and device-makers using AI as a hammer in search of a nail. 
I hope I’m wrong (and, let’s be honest, I probably am), because I’d love a new reason to get truly excited about Apple or Google talking about AI during their big unveilings. Until then, I expect to hear a lot more about better battery life, and even better photos and speech recognition. And while they’ll continue to be useful improvements, I also expect the novelty of “we’re doing AI in your phone!” will wear off pretty quickly without some new tricks to back up the claims.

Sponsor: Bonsai
Sponsor: Bonsai
Artificial intelligence
This is a really good post from Jeremy Howard of (check him out on a recent podcast interview) about the promise of deep learning in medicine, and also how to make it become reality. He places a big emphasis on data gathering, cleaning, etc, which is of course the less-sexy but very necessary part of the solution.
Yes, this is a case study from Google Cloud, but it provides some good insights into why a company like Box would be interested in AI. It’s not the most glamorous application, but there’s a lot of content out there just waiting to be labeled and, perhaps, used again.
Kai-Fu Lee of Sinovation Ventures has a lot to say in this interview, mostly about how AI will take pretty much all jobs from banking to plumbing. There’s good evidence on both sides of this debate.
This kind of underscores my point about Apple’s new AI chip. It boosts the performance of object recognition algorithms in surveillance video, which is good, but it doesn’t enable anything new. 
This is the first I’ve heard of this. It’s an interesting problem, and also AWS is giving away a bunch of cloud credits (and a little cash) to the top 100 teams.
TensorFlow Agents help researchers train reinforcement learning models faster by doing both simulation and processing in parallel. Speed will matter a lot as these techniques make their way into the real world and users want to get good results without wasting too much time.  •  Share
Sponsor: DigitalOcean
Sponsor: DigitalOcean
Cloud and infrastructure
This is just one GPU type on one specific AWS instance, but it’s the latest in a series of wins for AMD among cloud providers. Maybe they can help its latest comeback be for real.
China might be the epicenter of the server business right now, as it’s building and selling the only things of which companies are buying more. Given Huawei’s big spike, I’d guess Chinese companies are buying a lot or servers, too.
It seems like this is only for testing purposes but, still, free is free—and a great way to perhaps get folks seriously thinking about ARM for data center workloads. On a side note, I spoke with Packet CEO Zach Smith for the podcast last week, and that should air on Thursday.
This would be less interesting if it weren’t about essentially the same thing and VMware and AWS announced recently. IBM and VMware are claiming a lot of enterprises have already ported VMware workloads to IBM’s cloud, which would suggest AWS won’t suck all the cloud revenue from VMware users.
If you enjoy the newsletter, please help spread the word via Twitter, or however else you see fit.
If you’re interested in sponsoring the newsletter and/or the ARCHITECHT Show podcast, please drop me a line.
Use Feedly? Get ARCHITECHT via RSS here:
All things data
Qubole is a platform for running big data workloads in the cloud. And while its new “autonomous” features include a lot of suggestions about the actual data, the coolest part might be the suggestions about how to optimize AWS Spot Instances for your particular workloads.
Tools for getting people to share data and models across the company are hardly new, but they continue to proliferate. Maybe the problem is more cultural than technological.
For the DIYers and startup types out there …
New ARCHITECHT Show every Thursday; new AI & Robot Show every Friday!
New ARCHITECHT Show every Thursday; new AI & Robot Show every Friday!
Did you enjoy this issue?
ARCHITECHT delivers the most interesting news and information about the business impacts of cloud computing, artificial intelligence, and other trends reshaping enterprise IT. Curated by Derrick Harris. Check out the Architecht site at
Carefully curated by ARCHITECHT with Revue. If you were forwarded this newsletter and you like it, you can subscribe here. If you don't want these updates anymore, please unsubscribe here.