Is AI Practical for commercial Applications?
Phil Boyer [1:01 PM]
Howdy everyone! I’ll be kicking off the discussion here.
Josef Feldman [1:01 PM]
@jason @philboyer will be leading a Data Chat shortly.
The topic is — is AI really practical for commercial applications?
Please read: https://www.technologyreview.com/s/601139/how-google-plans-to-solve-artificial-intelligence/
Participants include:
Dennis Mortensen- CEO and Founder, X.ai
Sandy Steier- CEO and Co-founder, 1010 Data
Naveen Selvadurai — CoFounder, FourSquare
Ann Miura-Ko — Founding Partner, Floodgate
Sheila Gulati- Managing Director, Tola Capital
Shivon Zilis- Founding Member, Bloomberg Beta
Alex White, Co-founder, Next Big Sound
Amit Karp — Vice President, Bessemer Venture Partners
Jesse Beyroutey- Partner, IA Ventures
Ross Fubini- Partner, Canaan Partners
Mark Johnson- Co-Founder and CEO, Descartes Labs
Jason Black- Analyst, RRE
Phil Boyer- Associate, Crosslink Capital
Sven Kreiss — Lead Data Scientist, Wildcard
Jim Hao- Associate, Firstmark
Morgan Polotan- Bloomberg Beta
Scott Clark- Founder, SigOpt
Sage Wohns- Founder, Agolo
Nikhil Srinivasan- Founder, Distributed Systems
Peter Brodsky- Founder, Hyper Science
Tim Porter- CEO and Founder, Gluru
Phil Boyer [1:03 PM]
Topic: Is AI really practical for commercial applications?
While this is the official topic, I think a good place to start is to ask a couple more specific questions: When is AI practical vs. overkill? Is it practical for all enterprises? What are some interesting areas where AI is already making a lasting impact? When is AI NOT worth the investment?
Jason Black [1:04 PM]
@philboyer: great clarification
[1:05]
I think the upfront investment in AI for most enterprises doesn’t make sense at this point as AI startups fill the gaps
[1:06]
Especially if the enterprise isn’t already heavily leveraging data science and machine learning
[1:07]
And as increasingly larger companies get more comfortable with smaller and smaller software providers, it creates a huge opportunity for startups to sell into those accounts
[1:08]
that being said, it depends on how closely the 3rd party solutions fall to each enterprises’ core competency
Phil Boyer [1:08 PM]
However, Salesforce recently announced 3 acquisitions of AI companies (seemingly acquihires) in order to help develop applications. Do we think this will be a successful strategy for them? Or will the startups win?
Jason Black [1:11 PM]
Historically Salesforce has used acquisitions to build out their product offerings. Heroku, Desk.com, RelateIQ, Exact Target, etc. are great examples there. They haven’t done a ton to use that new talent to augment their existing products to date — though that can always change
[1:13]
Continuing on the earlier discussion, I feel like there are also going to be tools out there that just make it easier to build in house as well, thereby lowering the upfront investment
[1:13]
Tools like prediction.io, h2o.ai, or platform providers like databricks
Phil Boyer [1:14 PM]
Fair. I think the technologies (open source, commercial dev tools) will lower upfront investment. But do we have examples of companies or industries where the investment is showing positive ROI yet? Still too early to tell/
[1:14]
?
Jason Black [1:15 PM]
Would love to hear from @peter @prashant @sagewohns @sheila or @markjohnson who I see are in the caht here…
Phil Boyer [1:16 PM]
@markjohnson started a company on the premise that deep learning is a worthwhile investment. Would suspect he has an opinion here…
[1:17]
#callout
Jason Black [1:19 PM]
Well, I think the continued growth of data teams within enterprises is a proxy for an ROI
Josef Feldman [1:19 PM]
I think breaking this into target markets is helpful — enterprise & Dev tools are significantly different from consumer applications.
Phil Boyer [1:20 PM]
@jason: Maybe. It is definitely a proxy for “I” — not necessarily “R”!
Jason Black [1:20 PM]
I just spoke to an data lead at AT&T who said they’re leveraging data down to the individual truck and have determined that, due to the increase in traffic accident rates and decrease in speed, AT&T service trucks shouldn’t make left turns at intersections
[1:20]
which is pretty intense
[1:21]
@philboyer: that’s a fair characterization
[1:21]
but you’d expect data teams to stagnate or shrink if there wasn’t a justifiable “R”
[1:22]
though it’s an interesting observation that maybe the “R” is defined not by $ but maybe by some other metric
Phil Boyer [1:22 PM]
Yes — and it depends on what enterprises feel is an appropriate timeline to seeing the returns on investment
Jason Black [1:22 PM]
just as much Android adoption doesn’t directly affect Google’s bottom line, but it does increase their likelihood that a particular user uses Chrome to search the web (and spend more time on the web in general)
[1:22]
also very true
[1:23]
I think at a certain point, though, it will be too costly *not* to leverage data and ML
[1:23]
and maybe the investment today is driven by that expectation
Phil Boyer [1:24 PM]
True. I think there is starting to be a “keeping up with the Joneses” attitude in enterprise adoption/hiring in data science / ML
Jason Black [1:24 PM]
exactly
Phil Boyer [1:25 PM]
They prob don’t know what the F they are doing though
Jason Black [1:25 PM]
Gartner actually recently put out a report on Hadoop that highlighted a lot of the challenges
Morgan [1:25 PM]
no different than adoption of software and the web in the 90’s and 00’s was “keeping up with the joneses”?
[1:25]
i see ML as eating software, which is eating the world
Jason Black [1:25 PM]
uploaded and commented on this image: gartner_hadoop-100585449-large.idge.png
1 Comment
This doesn’t bode well for Hadoop’s biggest vendors. After all, as Gartner analyst Nick Huedecker posits, “Hadoop [is] overkill for the problems the business[es surveyed] face, implying the opportunity costs of implementing Hadoop [are] too high relative to the expected benefit.”
Jason Black [1:25 PM]
^ source http://www.infoworld.com/article/2922720/big-data/hadoop-demand-falls-as-other-big-data-tech-rises.html (edited)
[1:25]
ML = software
[1:26]
But I get what you’re saying
[1:26]
there are a lot of deterministic, rules based programs out ther
Phil Boyer [1:26 PM]
@morganpolotan: I agree — I think that ML is an extension of where software is today and AI the next extension.
Morgan [1:26 PM]
@philboyer: how do you differentiate ML and AI?
[1:26]
line always seemed blurry to me
Jason Black [1:26 PM]
those will likely start to leverage more subtle decision making engines enabled by AI
Phil Boyer [1:26 PM]
the question is really is investment into AI seeing returns yet?
[1:28]
The way I think of it, ML enables a program to select the appropriate algorithm to help arrive at a desired outcome. AI is essentially modeled after the brain to determine what the desired outcome is, which can constantly change based on a multitude of factors.
Jason Black [1:28 PM]
added and commented on this Plain Text snippet: ML v AI
“Machine learning deals with designing and developing algorithms to evolve behaviors based on empirical data. One key goal of machine learning is to be able to generalize from limited sets of data (paraphrased from [1]). Russell and Norvig [2] lists machine learning as a specific capability, namely the ability to “adapt to new circumstances and to detect and extrapolate patterns”.
Artificial intelligence encompasses other areas apart from machine learning, including knowledge representation, natural language processing/understanding, planning, robotics etc.”
1 Comment
Just to answer your question, @philboyer
Morgan [1:30 PM]
love that definition
Sage Wohns [1:30 PM]
that’s a great definition @jason, I see AI as a ML stack that able to emulate previously human-exclusive properties
Morgan [1:30 PM]
people use both terms interchangeably
Sheila [1:31 PM]
I’ve seen a lot of companies that provide data-centric outcomes (using ML or AI) for enterprises (and govt) and the results can be extraordinary. Often better than the results we would see from an internal data science team initially. I’m very positive on ML and AI companies driving real business outcomes (and then becoming a feature of every software company ever time). I do believe we will see greater productivity of internal data science as well especially as the third party platform tech and tooling gets better. The opportunity is so vast that there is clearly room for all (bespoke solutions, horizontal and vertical AI and ML, internal data science enablement and teams)
Phil Boyer [1:36 PM]
Agreed @sheila. However when I see a company in the space, I want them to focus on the best solution for the problem they are solving, not necessarily the technology at the forefront. Using ML or AI as your software approach seems to make the most sense when you have a massive, high quality dataset that continuously improves your application. in other cases, it may not be worth the investment in infrastructure, talent, etc.
Jason Black [1:39 PM]
Returning to the initial question around when does it make sense to build/leverage ML within an enterprise, there’s probably some way to categorize the stages companies take on the path to data/ML enlightenment. Maybe something along the lines of: identifying data sources, data exploration and transformation, data insights, data/ML driven decisions, formalized data/ML analysis pipeline/mechanisms, data integrated into product/core tooling, data enlightenment
[1:41]
Any thoughts on that path?
[1:42]
Feel like most non-tech enterprises are probably hovering around “data exploration and transformation”
[1:43]
Hence the large opportunity to step in and provide tools/products/services for things much further down the chain
Phil Boyer [1:44 PM]
ETL and data cleansing seems to be the area where most enterprises still get hung up. Hard to scale this appropriately when you have a lot of disparate data sources and formats
Jason Black [1:46 PM]
yeah, maybe it makes sense to break that out into 2 steps then: first “data transformation/cleansing” and then “data exploration”
Phil Boyer [1:47 PM]
Alright folks — have to sign off. Some great stuff in here! thanks for joining us
Morgan [1:47 PM]
great talk!
Jason Black [1:48 PM]
Same. Thanks, everyone!
[1:48]
@josef_feldman & @benjaminzeitz will be in touch about the next round
Josef Feldman [1:49 PM]
Thanks guys! @philboyer @jason @benjaminzeitz @channel