jkestelyn

They don't have to be awful; most of them are just poorly designed. Another Area 120 startup, Chatbase, is a service for designing voice bots that are much smarter and versatile than the norm by anticipating what questions people ask based on precedent (and all the different ways they ask them). Most voice bots can't do that.

[Disclaimer: I work on the Chatbase team]

"But researchers must also deal with the unexpected"

You'd think from this article that deploying a bot is like letting a dog off its leash and you just have to hope that it doesn't bite anyone. In fact there are several common-sense ways to prevent a bot from going rogue, including monitoring how it responds to intents and then making changes as needed.

So analytics are only useful for successful techs with wide adoption? I think many of those techs got that way because of analytics.

I don't disagree but how is that relevant?

"The technology to make AI-powered assistants truly useful is still far out of reach, and people aren't rushing to close that gap by adapting their behavior."

Exactly; if you build a bot with the intention of relying 100% on NLP, you're asking for trouble. But I fail to see how that fact leads to chatbots being useless when there are plenty of tools available for guiding necessary optimizations that can lead to a much better experience.

When websites were bad we turned to tools like Google Analytics to make website development data driven -- we didn't stop building them.

We value and appreciate those first VoiceLabs users who have made the jump to Chatbase; we're committed to ensuring you don't miss a beat!

I don't disagree; my suggestion was tongue-in-cheek. There is no value in optimizing an application that is so flawed that it shouldn't exist to begin with.

Those flaws derive from a wildly optimistic use case for the technology, though. A much cleaner use case would have been a bot intended for Facebook Help (instead of, or to complement, a KB -- assuming people still need that).

More ambitious maybe, but perhaps not impossible, would be a bot that looks for signs of suicidal tendencies in posts or comments and engages the user in therapeutic conversation. (?)

Leaving the clickbait-y headline aside, this story is really problematic because it conflates technology with an application of that technology.

1. When reasonably scoped (i.e., to a specific use case) and iteratively optimized over time, chatbots can meet user expectations quite well.

2. If ostensibly intended by their builders to handle every type of request on the fly via ML pixie dust, chatbots can be miserable failures.

Both things can be equally true.

(Disclaimer: I work for Chatbase, a service for analyzing and optimizing bots. Maybe Facebook should have looked at that. :) )

Sure, but past performance is not a predictor of future results. Just because many of them are built poorly today doesn't mean they have to be built that way.

Building bots is hard: there are few tools available of any sophistication, leading to a vicious cycle of trial and error (with users getting the worst of it).

Remember what the first websites looked like? Everyone was just making it up as they went along. We're in a similar situation today.

"Your chatbot should be purposeful, reflective of your product’s voice, and simpatico with your users"

Agreed 100% but IMO, this is not a function of "personality" but rather a function of deeply understanding user intents. A bot cannot be purposeful if its own designers don't know its purpose from a user-centric perspective.

(Disclaimer: I work on Chatbase, a service for analyzing and optimizing bots)

Relevant to this topic: Description of exactly-once implementation in Google Cloud Dataflow + what "exactly once" means in context of streaming:

https://cloud.google.com/blog/big-data/2017/05/after-lambda-...

(Google Cloud emp speaking)

GCP offers a $300 credit that expires after 1 year. It's a good way to get your feet wet with Hadoop and Spark via Cloud Dataproc. (Google Cloud employee speaking.)

Reminiscent of Walker Percy's essay "The Delta Factor", in which he theorizes that the essence of human-ness derives from the "linguistic triangle" (thing + word + human brain).

Based on the author's misunderstanding of AI, it has not "arrived" and probably never will.

If in the future, authors of such opinions would just let this simple concept sink in first -- that in machine learning application behavior is deduced from data rather than from fixed rules, but that in both cases the boundaries are set by humans -- we'd all be better off because their wild Skynet takes would never see the light of day.

As usual, I am more worried about the humans than the machines.

Warms my heart to hear that!

"Alien boys, rock steady"

Another huge fan here since the Black Flag days.

Rollins did a book of poetry in the 90s called "See a Grown Man Cry"...had a profound impact on me, and I've never run across anyone else who read it.

There are numerous historical examples of societal manipulation (of the most extreme kind) that had little to do with technology.

I don't disagree about the need for policies; I just think those policies need to be directed toward humans as the weak link in the chain, not machines. This is not like gun control where a single person with a weapon can do a lot of unchecked damage, and thus access to guns needs to be controlled. Very few have access/can do damage with an algorithm.

"It can be expected that supercomputers will soon surpass human capabilities in almost all areas—somewhere between 2020 and 2060...Is this alarmism?"

Short answer: Yes.

What is the evidence for this claim, upon which the premise of this article rests? The existence of algorithms that simulate human game play today is hardly it.

The authors have fallen into the trap of accepting the nomenclature of "artificial intelligence" without further questions. There is nothing "artificial" nor "intelligent" about it.

Rather, machine-learning algorithms are trained on a diet of human-derived data that is simply a reflection of existing human biases. The danger is in their human programmers being non-introspective of those biases, not in the algorithms themselves. Thus I personally am much more fearful of human-made decisions than non-human ones.

Don't hate the player, hate the game.

BTW, worth mentioning that Martin (the author of this tutorial) will present it in person at Google Cloud NEXT [1] in SF on March 8.

[1] https://cloudnext.withgoogle.com/schedule#target=tensorflow-...

IMO, the same question could be asked with respect to any abstraction that hides complexity. The answer will always be/has always been: No, it won't. Rather, engineers who can think creatively about the role of pre-built ML models in mainstream applications will be in high demand.

The 1980s called, they want their expert systems back.

Really good write-up (about a year old) on precisely that subject, here:

https://sematext.com/blog/2015/01/30/solr-elasticsearch-comp...

All the angst about "big" data that "isn't big" is based on a false premise. "Big data" was NEVER just about scale, but rather intended to be equally descriptive of diversity as well as velocity. The problem is that whomever coined the term made the same strategic error as whomever coined "global warming" -- the adjectives used are too specific to adequately describe the full range of qualities involved.

Just with respect to some details: Spotify changed from being a self-supporting (free) user of Cloudera's open source platform to being a paid support customer of Hortonworks. So, it was never actually a Cloudera customer (despite what was widely "reported").

Disclaimer: Cloudera emp here.

More technical details available here:

http://blog.cloudera.com/blog/2016/02/introducing-apache-arr...

PMC Chair = Apache VP

The flip side of this survey: 46% have either deployed Hadoop or plan to deploy. That's a pretty healthy number and qualifies as "mainstream" adoption, IMO.

Very nice effort. A couple things missing though to make it truly up to date:

1. In "Ingest", where's Flume? 2. Where's "Interactive SQL" (eg Impala, and for Presto)? 3. Where's "Search" (Solr, ElasticSearch)?

Also of interest:

"Algorithms Every Data Scientist Should Know: Reservoir Sampling" http://blog.cloudera.com/blog/2013/04/hadoop-stratified-rand...