Is Uber downsizing a proof of AI hype?

Amazon
bejeezos

Go to company page Amazon

bejeezos
May 18, 2020 309 Comments

As you may have heard Uber is cutting 3000 additional jobs, closing 45 offices, and reevaluating many of it's bets. One thing that stood out to the Artificial Intelligence community is the statement that Uber doesn't think AI research is core part of their business.

"As part of the latest changes, Uber will scale back on noncore businesses. Mr. Khosrowshahi said the company is winding down its product incubator and artificial-intelligence lab, and exploring “strategic alternatives” for Uber Works, which pairs prospective employers with gig workers."

Source: https://www.wsj.com/articles/uber-cuts-3-000-more-jobs-shuts-45-offices-in-coronavirus-crunch-11589814608?redirect=amp#click=https://t.co/J9owqvEdjV

Do you think this is a precursor that we're living in an era of overhyped promises about what AI can deliver? Are AI science team more expendable than engineering teams?

4788 PARTICIPANTS SELECT ONLY ONE ANSWER
VOTE VIEW RESULT

comments

Want to comment? LOG IN or SIGN UP
TOP 309 Comments
  • It got bad because in the past 5 years every college grad is am “ML enthusiast” and that killed the potential. Now they will run and hide - at some point they were thinking it’s the best way to earn more . (I am in ML and I see how bad the talent is)
    May 18, 2020 16
    • Amadeus
      JnDH67

      Go to company page Amadeus

      JnDH67
      @Xometry soon the AI/ML jobs will be over saturated with people.
      May 20, 2020
    • Xometry
      tSBA86

      Xometry

      tSBA86
      @Amadeus Again there will be a huge difference between wanna be and someone who really has knowledge. AI is more of an art and research than learning and applying. I bet someone can learn those tools within 3 months but applying and actually solving a problem can take decades to learn unlike software engineering.
      May 20, 2020
  • NetApp
    love-of-TC

    Go to company page NetApp

    love-of-TC
    - I don't think Uber could rely on its collected data/talent to utilize AI for growth (beyond some core applications such as fraud detection and demand prediction). It's not as profitable as maximizing engagement time or ad revenue, for example.
    - Your suggestion that this article can be a proof of AI hype is somewhat cherry picked.
    - There is a wide misunderstanding of what can be achieved with data, and many marketers ride the "AI hype", which pisses off dedicated researchers who now have to compete with garbage research work flooding the field and creating just noise and damage.
    - It's easy to build *something* with AI (e.g. image classifier - just fork from GitHub!), but requires much more to solve a *specific* data dependent problem and deploy a probabilistic solution that can reliably cover unseen cases as well. Very different from engineering, heavy maths foundations are crucial.
    - Startups and companies that build effective ML solutions can solve an entirely new realm of problems at scale. There is a tangible and irreplaceable value from some AI applications.
    - What indeed stinks is companies insisting on using AI even when unnecessary. You see miserable graduates forcefully applying decision trees to business problems that aren't properly defined or can't scale, just to have their managers' satisfactory boasting to use AI.
    May 18, 2020 7
    • Uber / Eng
      🕴🏿⚰️🕴🏿

      Go to company page Uber Eng

      PRE
      Google
      🕴🏿⚰️🕴🏿
      @Grubhub what is dysfunctional? If you are referring to OP's, what does he know? Uber is extremely bloated but isn't dysfunctional. The large scale products don't depend on one bloated research department at Uber. If NetApp has any idea what it takes to build large-scaled system, he would have known what role research scientists play. He sounds like some first-year graduate student who read few papers and thinks he understand the engineering work around large-scaled ML system.
      May 18, 2020
    • NetApp
      love-of-TC

      Go to company page NetApp

      love-of-TC
      @Uber I fixed the first bullet of my post. No intention to claim that Uber misuses AI/lacks talent. I actually spoke to ML guys from Uber and following both Uber AI labs' research work and the ML platform engineering publications (e.g. Michaelangelo).

      If you're looking for more opinions (addressing mostly Uber researchers and false promises of autonomous drivers) check reddit:
      https://www.reddit.com/r/MachineLearning/comments/gm80x2/n_uber_to_cut_3000_jobs_including_rollbacks_on_ai/
      May 18, 2020
  • Uber CEO is a finance guy and not a tech person. He got lucky with Expedia in a thriving economy and is getting tested for the first time now. It doesn't seem like his vision aligns with that of a tech company of the future. This ruthlessness and nonchalance is something you'd see on wall street.
    May 18, 2020 6
    • @JoshBrown, you've got me wrong. I am not advocating for over-expenditure. You're spot on about the way tech companies are allowed to go public in the US without making a dime in profit. It's all about increasing the revenue by pumping in VC money and these VCs shift the burden from themselves to the small-time retail investors who know nothing about the company fundamentals or its future outlook. That is a more fundamental problem and there needs to be a gigantic paradigm shift in the US economy in order to address that issue. In fact, I agree more with your point than mine keeping the larger picture in mind.

      However, my comment is simply related to Uber's woes in the current climate and why it is faring worse than any other tech company of its stature. This seems to me like a lay-off parade being conducted to distract people from the core issues at the company and the CEO's inability to stabilize the ship. It is an excuse to secure the CEO role at the cost of those who've built the product and the company from the ground up, way before this CEO came on board, which is simply unacceptable and immoral. I am not a fan of Travis Kalanick but I can vouch for the fact that he being the one who founded this company had a solid vision and an idea as to where it's headed which the current leadership seems to be lacking entirely.
      May 18, 2020
    • Live Oak Bank
      Djkrs4533

      Live Oak Bank

      PRE
      McKinsey, BCG
      Djkrs4533
      @Google, yes that makes a lot of sense. In order to get taxi cheap enough, you need L5, and that's what Waymo could bring to Lyft, though I still think that takes 5+ years, but Waymo can get a ton of data from Lyft ( though they could just invest in Lyft and get the data)
      May 18, 2020
  • Uber
    vMNw42

    Go to company page Uber

    vMNw42
    AI labs at Uber is specifically research focused with no connection to product / the business. The company needs to become profitable so research roles are just nice to have at this point.
    May 18, 2020 0
  • Uber
    oxpT23

    Go to company page Uber

    oxpT23
    The AI Labs team was a research group of ~20 ppl focused on academic research with occasional applications to the business. I tried to work with them twice on business applications, and they were pretty much useless. Uber still has several hundred data scientists and machine learning engineers involved in building and maintaining several hundred machine learning models. You would be surprised at how complicated a "taxi app" can be.

    After all, why does Google Search need so many people, it's just ranking websites and selling ads? I read a paper on it once, so it sounds pretty solved to me...
    May 18, 2020 6
    • Google is complicated is because of scale. Facebook is complicated because of scale. AI at those companies is also complicated because of scale.

      But AI is still hyped in general. No one is making money out of pure play AI but the big guys.
      May 18, 2020
    • Uber
      oxpT23

      Go to company page Uber

      oxpT23
      @bapbo if you survive long enough to attempt building a taxi network, you'll understand why 200 engineers building "the core" will not be enough.
      May 18, 2020