MIT and IBM establish joint AI research lab | Hacker News Reader

MIT and IBM establish joint AI research lab

news.mit.edu

86 pointsffwang227 comments

IBM bought visibility with its ad campaigns, and it's trying to buy credibility with this sponsorship, but I'll bet you anything that the only part of IBM that MIT will use is the brand. The tech will be open-source tools that IBM didn't produce, and it certainly won't be Watson.

Yes, "Watson" is really old technology coupled with costly consulting services. And yes, IBM is far behind Google, Facebook and other Silicon Valley companies in AI research.

That said, for MIT, which in some ways is trying to catch up with schools like Toronto and Stanford in AI research, particularly in deep learning, it makes a lot of sense to take the $240 million from IBM to create a dedicated AI research center.

I think they have a shot at becoming an important 'center of gravity' for AI research in the East Coast.

Well MIT CSAIL is an acronym for AI Lab :) and it was founded in 1959. Here is an old link which lists schools particularly strong in different domains of AI - https://www.cs.cmu.edu/Groups/AI/html/faqs/ai/old_ai_general...

In some ways, MIT needs to catch up. For example, here's a list of deep learning research groups put together by Bengio's group at Montreal as of Oct 2016: http://deeplearning.net/deep-learning-research-groups-and-la... -- MIT is missing from the list.

But today anyone can ask IBM for some Watson queries. Which AI services can I buy from Google, Facebook and others ?

Voice, image, video, text analysis, translation:

https://cloud.google.com/products/machine-learning/

> But today anyone can ask IBM for some Watson queries

That's the metric for cool AI functionality? The Watson query tool is basically a search engine with a layer of (not super reliable) text synthesis over it. It would be more interesting if they took away everything except the document search and passage extraction stuff, and just presented that.

This is one of the many desperate moves that we will see from IBM trying to deliver on the Watson overpromise and all their rocambolesque cognitive computing. AI is a late project (https://en.wikipedia.org/wiki/The_Mythical_Man-Month). So adding more manpower (or brainpower) to a late software project only makes it later. Only folks poorly educated on the topic (such as Elon Musk) really believe that Watson-like AI is around the corner.

I don't think it's universally true that adding more manpower to a late software project always makes it later. The Mythical Man Month takeaway is that things don't scale up linearly when adding manpower but they clearly can and often times do scale up.

Imagine one person given the task of rewriting the Windows operating system in Java. This one person team won't benefit from more programmers? Of course the project will be completed faster, in this case, with more manpower.

> I don't think it's universally true that adding more manpower to a late software project always makes it later.

It always results in a period of reduced progress due to drag on the existing staff to onboard the new staff; and the bigger the scale up, the longer that period where you are behind where you would have been without it is. And the more you scale up, the more you need to reorganize and build new coordination infrastructure to make use of new resources even once they are up to speed technically, which also takes time to set up and time to acclimate staff to the new organization and teams, which creates its own drag.

In realistic scenarios, this pretty invariably means late project + more resources = later projects.

But, sure, there are extreme situations where that wouldn't be true, but I don't think they pop up often in practice. One should be extraordinarily skeptical of any claim (or interior intuition) that the rule doesn't apply to your project.

What I presented was an extreme example but the point stands. Obviously things like Windows OS, iOS, etc. are projects that benefitted from an increase in manpower and this increase in manpower did not stifle development or cause delays. It is true that doubling the manpower does not double progress. It's not a linear relationship.

rocambolesque adj.: Suggestive of Rocambole, a character in the novels of the French author Ponson du Terrail (1829–71), renowned for his improbable and fanciful adventures; incredible, fantastic, bizarre.

https://en.oxforddictionaries.com/definition/rocambolesque

I've never believed that Musk could be so daft. I always thought he was playing up media hype ('AI so powerful, could be evil guys!') as a way to recruit young ambitious talent to his endeavors.

I use to collaborate with IBM Research. Basically anyone worth a damn eventually goes to Google/FB/MSFT. IBM pays less than half of their competition and doesn't provide a fraction of the resources. Their only appeal is that they have a lower bar for the "research scientist" role, which appeals to a lot of PhDs who may not have enough top publications to their name.

You really have to give IBM credit for their marketing.

Then again, it's always going to be this debate of engineers clamoring for better engineering and consultants clamoring for flashier powerpoint decks, animations, and other gimmicks.

Good technology (executed with an end-business goal in mind) delivers real value, and that real value is what will sell in perpetuity. Otherwise, you can only go on fooling people for so long...

I would be curious to see how the MIT academics play with the IBM consultants / engineers. Would be curious to hear anyone's comment on the inside of this closely connected.

> I would be curious to see how the MIT academics play with the IBM consultants / engineers.

Almost certainly, the IBM engineers they play with won't be the typical contractors customers normally see. They'll be high-quality and knowledgeable.

This is super interesting. I know there is much heated discussion here on HN about IBM in terms of the Watson project and some of the claims made vs. reality etc. I am pretty far removed from the trajectory of IBM's AI research, but can only figure a joint effort with MIT will bolster their reputation by association if nothing else.

Just finished reading about Facebook and Microsoft launching a joint AI effort not five minutes ago. To those enmeshed in or working in AI research: are these joint efforts a response to Tensorflow and Google?

My cynical take: For a company like IBM it doesn't take a lot of money to sponsor an institute. It's likely a marketing effort. They've seen all of the bad press.

Indeed. $24 million a year is pocket change compared to the advertising budget of Watson. Additionally, if it nets them a little academic credibility and a few MIT student grads per year, all the better.

I think the Facebook + Microsoft thing is definitely a response to TensorFlow. Google seems to be positioning Keras as the portability layer above frameworks and frameworks are implementing it too.

This seems more straightforward in that IBM has money and MIT has a good reputation and also a lot of smart folks that they may be able to hire.

All due apologies to the late Marvin Minsky, but has MIT really been a leader in AI lately?

No, and that's why they're taking this deal.

Got to wonder why MIT would shackle themselves to the albatross that is Watson. Or given Google's antics with think tanks, be willing to take the risk that unfavourable results will be career-limiting.

240 million reasons. No doubt MIT will mention IBM in their reportage of this collaboration. But I bet they use the word "Watson" as little as possible.

Is this only going to be at IBM's new facility on Binney or are they going to have space in the facilities MIT is building now?

I wonder if it's a good thing that the new norm is for entire AI research labs to be gobbled up by giant tech companies.

You do need a lot of computational power and sufficient test bed in the long run. You certainly can fo AI research on a dozen computers, but would be far better if someone else handle the infrastructure and you just do the research. It makes sense to partner with the giant tech. Also, well-known researchers almost always do joint research with researchers at giant tech because they know each other already.

Reply on news.ycombinator.com