I am actually shocked to learn that x.ai has human beings reading the emails.
X.AI's marketing materials state it's "An AI personal assistant who schedules meetings for you."
Their default tagline in every email sent says "x.ai – artificial intelligence that schedules meetings "
You have to dig into their press kit to get any mention of human "Supervised Learning".
My typical interaction with Amy has been a 3rd party suddenly CC'ing her (it) into an existing thread.
In many of those cases those threads contain information that I would consider to be confidential.
In some cases the people on the other end work for public companies that I am POSITIVE would not allow for a non-approved human to have access to that information.
I understand the counter argument that x.ai's warehousing this info regardless, but introducing a human under the guise of a blind-AI is unsettling.
It's a disingenuous pitch and should have repercussions if they don't improve their disclosures of what the service is actually doing.
"x.ai – artificial intelligence that schedules meetings"
Reality:
"A group of low-paid humans (that may-or-may not have been background checked) will read your emails to help you schedule meetings. They will probably not use this information in any way other than intended."
Those feel like two different products that I would make fundamentally different decisions about.
Yes and I think that works with many areas and in many products.
But when you have confidential information in the mix - especially stuff that might have SEC implications - it changes the game.
A few years ago, I did a contract with a company that had a system that deleted all email greater than a year old. While the official answer was that it was "to save space and improve network performance" I suspect the unstated reason was to prevent fishing expeditions.
If your email is being cc'd outside the org and read by actual humans, that introduces some awkward problems.. and may force people to admit the actual reason for the policy. ;)
Meekan is a Slack bot that's actually 100% AI, takes natural language commands, and is used by 10's of thousands of real people. The Median time to schedule a meeting with Meekan is 53 seconds, which means he's delivering real value.
Not trying to be an AGI, the bot looks in everyone's calendars and prefs and suggests the best time to meet. By that, he's trimming the huge decision tree into a small, ranked suggestion list that's easy to grasp and decide upon.
An assistive tool that saves tons of time and frustrations. Nothing less.
Just curious, but why do you refer to it as "he", especially in this context where you are touting it as an actual non-human? Not to single you out, I see people referring to all these other things as (usually female) "her" and it just unsettles me to think that people are anthropomorphizing these things so much.
On a related tangent, often these same people will refer to animals as "it", too, which is even more curious. A living being with an actual gender doesn't rate a gendered pronoun, yet a piece of software does.
True. We started out trying to limit the conversation to just scheduling-related topics. But very early on we realized that since this is all happening in a conversation, people expect him to be able to carry some smalltalk - because when he doesn't, he's perceived as stupid - he can't even say "hello" properly, how can he possibly do scheduling?
(if this is interesting, I wrote about it a few month ago in much more detail: https://medium.com/building-the-robot-assistant/cheating-on-... )
I wonder how many other "cloud" companies have stuff that users assume are one on one conversations or sharing, and aren't. Even something as simple as a dating site almost certainly involves employees reading private messages (a must because of scammers, spammers, and downright abusive users.)
If a third party is CC'ing Amy I would of course send her the unencrypted confidential information and think it's just an AI - they don't look at my data. What the hell.
It's completely disingenuous and is riding on the AI hype train. I've never once ran into a chat "bot" that wasn't human. It's ridiculous, completely misleading, and frankly waters down any real startups working towards that future.
> I've never once ran into a chat "bot" that wasn't human.
Well that's absurd. Siri? Eliza? There are multitudes of actual chat bots out there. They usually good to fool a good lot of the general population. But the tapestry unravels with just a little bit of effort if you know what to ask.
Two parties to an email conversation. If one person is using "Amy" and adds her to the conversation, the other person may not fully understand the implications. Especially since their marketing tries hard to blur the lines.
Even if that is the case this third party would then assume that the secretary is "with" the other person. Not some other person that no one have ever talked to or met or etc.
There's a huge difference between an employee monitoring the product to make sure things don't go wrong, and an employee paid to fill in for the main product. It's like a real-life mechanical turk.
I get that they are trying to use the human responses as a training set.
But the scope of these general purpose concierge services mean that what they are going after is damn near Artificial General Intelligence. At least in terms of capability, after-all ordering a burrito and locating an antique skull is a pretty broad capability range.
And at what point of accuracy are they aiming for for the systems to run on their own?
I guess they're thinking they can get pretty good at identifying the easy requests the system can fulfill at 99.9999% accuracy and then shift the harder requests to a human.
At that point it seems like they're just building a system that can distinguish hard requests from easy requests. Sure fulfilling the easy requests with AI is a feat in itself, but how many of those easy requests does it make sense to go through a third party? Why doesn't the burrito shop just set up it's own ordering bot?
Am I missing something?
The edge cases seem so broad that to sufficiently fulfill them would require something like AGI and the more approachable tasks of ordering pizza or burritos seem unnecessary of a third party.
I've said it before[1] and I'll say it again...the surge of interest in chatbots is premature. We're not even close to solving this problem, because yeah, it has to "learn" and "understand" language, which is equivalent to full intelligence according to some theories.
My guess is, for the foreseeable future, these services will fall into one of two categories:
1. Backed by legions of humans, and thus not that
interesting.
2. Trivial at best (ordering pizza with marginally less effort than before, a glorified collection of Slack plugins), disastrous at worst (Microsoft's Tay, the subject of my other post).
I guess one answer to my question might be:
They are trying to build an AGI.
But one would have to say that the relatively sparse training data consisting mostly of takeout orders, and the sweatshop like environment would be a strange way to go about attempting such a thing.
The solution is so blindingly obvious: You are creating AI not to replace humans, but to assist humans. If your humans can manage 50 emails per hour, you judge the quality of improvements to the AI on how any more emails per hour they can handle. This is from improving the tools humans use to completely eliminating the human from the equation in certain cases. Then your customers need to rate the quality of the service to see if automatically handled emails had the side-effect of reducing quality, which then needs to be weighed against the cost savings.
No one here will be shocked at the eventual investor lawsuits, but I was a bit surprised at the "Pain -- Solution" graphic. Unless "Michael" is very senior to "John", he is a complete asshole:
J: I'll be free to meet at these times...
M: Yeah whatever. I can't be arsed to look at my calendar, so talk to my secretary. Oh wait... I'm too cheap to hire a secretary so why don't you talk to this robot instead?
I get it, Lords of the Universe can make you work out the details with their administrative assistant. But then they actually have to employ that person. Is this product pitch aimed at "temporarily embarrassed executives" who envy that sad dominance game?
You've correctly identified the problem x.ai really solves: I want a secretary but I'm not that important.
The ridiculous thing is that doodle.com already solved this problem in a much less sexy way - just click the time you want. Problem solved in 10 seconds (vs minutes with this dumb bot).
Wow, your comment plus TFA's description of the working conditions for the "Heroes" (perhaps the most Orwellian nomenclature I've seen this month): I'm suddenly reminded of an SF short story I must have read decades ago (the title and author of which escape me) in which tourists visiting Vietnam have "fun" in a demented "Vietnam War" theme park, where they get to gun down screaming robot peasants. Inevitably, a gruesome discovery is made... it suffices to say, Benjanun Sriduangkaew would not approve.
In the future, will "I thought it was a computer program behind the interface" be something like "I was just following orders"?
Isn't it ironic that chatbots are touted as a more natural, conversational interface when they are basically CLIs? Back in the day GUI adoption over CLI was driven by conviction that desktop, drag and drop and point and click are more natural metaphors.
As usual, what's old is new again. Makes sense given that user interfaces and "styles" of software are just like styles in anything else (ie: fashion) -- they come and go and come back again.
Fast forward another 3 - 7 years and users will be wondering at the marvels of a chatbot where one can simply click on the picture of the action they'd like it to perform! Perhaps entire menus of actions will be presented, in horizontal or vertical strips inside the chat interface... these actions will of course be represented by hieroglyphs expertly designed to convey just what they do and no more, to as cross-cultural an audience as possible. Maybe we'll call them icons or something.
Fast forward another 3 - 7 years and users will be wondering at the marvels of a chatbot where one can simply click on the picture of the action they'd like it to perform!
You mean like the current functionality of WeChat, Telegram, Kik and FacebookM?
I generally like command line-esque chatbots. Then again I also like the command line and usually prefer it over a GUI. (I also think that emacs is the best UI ever so I'm perhaps not representative of the typical user.)
The problem with x.ai is that it attempts to replicate the annoying experience of dealing with someone's idiot secretary, not that it's text.
Sounds like some of these companies aren't even implementing the automation part.
> But she and another former employee, Alex Gioiella, said the only automated part of the service they saw was the occasional marketing text message.
You'd think that the human workers would at least get served up suggestions by the system if it were actually learning, to make sure the algorithm got "gold-standard" feedback on its dataset.
Especially requests like those for delivery food could be standardized pretty quickly, so that workers would only be handed situations where the "AI" (if there even was any machine learning involved) was uncertain as to what needed to be done.
> But usually, the Hero said, the requests were for pizza or Chipotle delivery.
Funny comment from a Facebook friend: "The meta-turing test: A human trying to convince another human that it is chatting with a computer program. Thankfully a human dumbing itself to chatbot level is far more doable than the opposite." https://www.facebook.com/groups/cyberpunkculture/permalink/5...
Is there any indication that there is actually light at the end of this tunnel, and not just a lot more tunnel? How much of this is just smoke and mirrors for the sake of product development, and how much is just to bilk investors? I read the entire article, and I don't get the sense that the author is totally sure either.
The company with the $9/month price point has to reach near full automation or die. The company with the $200/month price point could be totally manual until the automated systems catch up.
Recommended viewing: "The Devil wears Prada", Andy's first day at Runway.
One of the reasons I left x.ai was that I believed the leadership was deliberately misleading users, investors, and employees with false advertising and extremely inflated and skewed metrics. I am sure that the company is making great progress, because it is full of very intelligent people tackling an interesting problem. But the way they treated employees, customers and investors was unacceptable to me.
I suppose on its face it's not a bad idea: pay some people to train an AI agent. It's a bit annoying that any information about how it's working out so far is unavailable, but on the other hand, it's also rather revealing that it is. If it was working that well we'd probably have heard a lot about their tech by now.
The goal/intent/promise with these companies is always train AI to automate the process. This is good for both "buzz" and cost reasons long-term.
Companies describe the AI as two-fold: 1) actual AI responding to your requests and 2) Automation making human workers' jobs much easier, ie. a really good CRM or scheduling algo.
The reality is that high growth almost always means throwing more people at the problem under the guise of being "trainers" when in reality the technology only really caters to #2 (better automation) and the company is forced to pivot before ever really getting to #1 (AI handling the requests) either due to distractions or inability to actual deliver on the promise.
Just one of a great many points of correspondence between the present day and cyberpunk dystopian fiction. (Technically, I think Diamond Age is post-cyberpunk.)
Over at Naked Capitalism, the daily news rundown has a section titled The Jackpot.
What was dystopian about Diamond Age? While it wasn't explicitly utopian, the world is much closer to utopia than dystopia. Certainly better than current reality.
X.AI's marketing materials state it's "An AI personal assistant who schedules meetings for you."
Their default tagline in every email sent says "x.ai – artificial intelligence that schedules meetings "
You have to dig into their press kit to get any mention of human "Supervised Learning".
My typical interaction with Amy has been a 3rd party suddenly CC'ing her (it) into an existing thread.
In many of those cases those threads contain information that I would consider to be confidential.
In some cases the people on the other end work for public companies that I am POSITIVE would not allow for a non-approved human to have access to that information.
I understand the counter argument that x.ai's warehousing this info regardless, but introducing a human under the guise of a blind-AI is unsettling.
It's a disingenuous pitch and should have repercussions if they don't improve their disclosures of what the service is actually doing.