Bots Are Hot (1996)
wired.com
wired.com
Serious question; I completely fail to see why there would all of a sudden be a gold rush here.
* There hasn't been some shiny new technology introduced (open source machine learning, maybe? Messengers with APIs have existed for a long time)
* There isn't some massive consumer shift happening that's new (messaging has been around forever. It's been crazy popular outside the US for a long long time.)
* There isn't some crazy new business model everyone's chasing
* There hasn't been some crazy UX breakthrough that solves text interface problems (high learning curve, ambiguity about what the service can do)
I can, however, understand why Facebook and Google want people to build bots on their new platform, hence the submarine reference.
Any ideas?
None of these is bot friendly at all. Slack and Telegram have been a huge improvement for my daily automation needs.
That sort of counts as a shiny new technology or at least a consumer shift, given the increasing ubiquity of these assistant agents today (every smartphone now has at least one installed by default and so does Windows 10 by default; the Echo has been a surprising sales hit for Amazon).
It doesn't seem surprising to me that companies are re-looking at matching up typing and text with these APIs, and coming up with some interesting use cases, especially with better NLP here in the last year.
I'm not doing anything bot related, but I think it's a (reasonably) exciting space right now.
I'm beginning to believe Peter Thiel's quote of "Every moment in technology happens only once." When one company has great success, it doesn't mean that a whole sector will have great success; indeed, it's pretty likely that everyone else in that sector will fail spectacularly. But if that's true, the business model of many VCs (figure out the next big thing, and invest in the companies do it) is fundamentally flawed, and they should stick to providing working capital for businesses that have struck it lucky and already started to take off. But if that's the case, capital is a commodity, and it's awfully hard to justify 2 & 20.
There are hundreds of use cases I can think of. A doctor implements a bot on his/her site and it asks potential patients questions about their condition. It feels much more friendly than reading a FAQ. The same goes for a lawyer firm. Or how about a bank’s call center. Actually, every call center out there could be replaced by sophisticated bots. Or a bookstore that asks you to name a few of your favorite books and then makes recommendations. Come to think of it, the market here is huge. That’s why all the big players are moving in. Microsoft is already building APIs for their platforms.
As for Google, judging by the fact that they’ve invested shitloads of money in AI research, bots could really give them an opportunity to capitalize on that know-how.
Much of what we don't use people for these days is out of necessity. Microsoft can't afford to have a chat window on every Windows machine directly connected to a 1:1 support agent, but with bots, you can. Same goes for.. everything, heh.
Bots won't always be welcome on every page and piece of software, just like people won't always be welcome in every web page and piece of software - but there is definitely potential.
And the bot misdiagnoses or has a malfunction, leading to the patient not getting the correct treatment they required. Yikes.
vs. a regular doctor doing that? Which happens all the time, actually. Its really interesting how we expect computers to be far more accurate than we ever expect humans to be!
Of course, I'm being a little disingenuous: You can sue a doctor, you can't sue a bot.
But you can sue the company that made the bot, and the company that employed it. With humans the employer can often dump responsibility on the "fall guy" who made the mistake, while with bots you often have to (properly) consider any failure as a system problem.
iOS is still king for apps, but this move benefits the above players at the expense of Apple (Google seems very hungry to cannibalize the Play Store if it can make the bot store work).
Furthermore (specifically for Google and Amazon), if text-based bots can work out, they're both making a play for the Echo/Home market, bypassing smartphones devices entirely.
- Tech is ready for realtime natural language processing
- Developers are getting bots right.
- Plenty of APIs to build task-focused assistants
- VC money
- Popular demand
My guess is that the reward is being able to become the gatekeeper for what service gets to fulfill the customer's wish.
Trying to become that gatekeeper for search is a lost cause...Google has monopolized it.
So, the barrier to entry, while still high, is lower than search.
Amazon's echo, for example, gets to be a gatekeeper, without having to cut a deal with Google, or have Apple's iPhone in the way. Amazon fully controls who gets the business when you ask for a taxi to the airport.
The software was written by mostly teenagers and by all modern metrics was absolutely atrocious. It required constant maintenance, crashed all the time, and was riddled with bugs that would reveal your character as a bot and get it banned. But the exchange value for the virtual currency was so high it didn't matter, I was still making hand over fist.
I didn't write any of the software I used at the time but it's a pretty frequent daydream of mine imagining the literal millions of dollars I could of made were I able to go back and re-write that software knowing everything about software development I know now.
The beauty of bots is that they do not augment the productivity of the user like most software, they replace it with something that is a consistant multiplier better.
It was a once in a lifetime thing because tens of thousands of players were flocking back to the game because of new updates, and Jagex (the company behind the game) was selling out for an acquisition by Insight Ventures Partners, so they inflated their player base by vastly scaling back their anti-bot measures (mainly bans).
Huge supply, meet huge demand. The Jagex founders (the Gower brothers) came away with $100m for the whole ordeal and power to them the technical achievements of the game (a browser-based MMO written in Java 2001 - present) was nothing short of remarkable [1].
I think the bottom line is that there is often very juicy low hanging fruit that you can get to just by a minimal amount of knowledge and sheer force of will. The deeper knowledge and experience will give you a lot more things to think about which will help with edge cases, scalability, maintainability, robustness, etc; but while those things may be necessary for some projects, in many cases they end up bringing only small marginal gains.
I'm at the point in my career now where I'm actively cultivating beginner's mind to avoid being paralyzed by my knowledge of all the ways things can go wrong.
But yeah, in general thats absolutely true.
The upper bounds that existed was actually caused by payment processing. Paypal vehemently sided against the seller of 'virtual items' which led to locking of accounts. That's what did me in at least.
The margins though decreased over time as the demand for virtual currency went down. In 2012 the proprietor of the most widely used botting-client went to work at Jagex and in 2013 they released a patch known as the "bot nuke" which quite appropriately wiped out maybe 90% of injection-based botting clients (which is what we all used).
I actually poked my head back into the scene when OSRS came out. But three things make gold farming on the game pointless: 1) Jagex monitoring systems are much better. They can ban you when they want to ban you, they still don't (I imagine because they want to wait for your membership to renew, as well as keep numbers up), but they'll get to you eventually if you're a serial offender. 2) the margins are still really low. Throwaway accounts will make $0.40/hr at most. 3) the software quality is even lower then before, the clients are poor and the scripts are poorer. All the best scriptures from ~2011 are probably full-time SWE's right now who can't be bothered.
TSO had some ham-fisted money making multi player games that forced players to interact with each other in exchange for Simoleons.
One of them was a maze solving game for two players, where one player is lost in the maze and can see the local walls around them, and the other player can see an overall map of the maze without the player's position, and has to guide them around and figure out where they are and how to guide them out by asking them what they see and telling them which direction to move.
That required a lot of work for two people to do manually, so there was a big reward, but it was trivial and fun to solve mazes programmatically! (Plus it made cool bleeping and kaching sounds as it solved the mazes and printed money!)
I would run two TSO clients at the same time, logged into different accounts in different windows. The bot attached to both of them, then screen scraped pixels and injected events to repeatedly solve mazes by moving the player around until it identified where they were, solving for the shortest path, and bringing them straight home quickly by machine-gun clicking on the arrow buttons.
My housemate had a good eBay reputation, so as an experiment, we tried selling Simoleons on eBay for real currency via PayPal.
I could generate arbitrarily huge amount of virtual currency quite quickly. The bottleneck was selling it, and the problem was customer service.
The problem was that many of the customers were pouty temperamental 15-year-olds using their parent's eBay accounts, who would give scathing eBay reviews if their order wasn't delivered instantly, or they suffered some imagined slight.
And the other problem was that TSO just wasn't designed to make it easy to transfer large amounts of Simoleons from player to player.
You couldn't just "wire" somebody an arbitrary amount of cash via in-game email -- you had to show up on their lot and meet them at a specific real time, and suspiciously hand it over to them $1000 at a time.
There was another better way to transfer cash more efficiently than handing it over grand by grand, and that was with tip jars: You could fill a tip jar with $5000 using the pie menu with a couple of mouse clicks, and then the user could empty it the same way.
So when I had to deliver our first million Simoleons, I came up with a system where I'd go to the lot of the customer and meet then, then ask them to line up a bunch of tip jars in a row. I would then use bot macros to fill each tip jar one by one with $5000, while the customer would quickly empty them as I filled them up, and then we'd go back to the beginning of the row and start all over again, until we'd transferred the entire million Simoleons, in only 200 $5000 hand=>jar=>hand transactions instead of 1000 $1000 hand=>hand transactions.
One time when we were making a big delivery of cash, running the gauntlet of tip jars in our customer's living room (which I admit looked pretty fishy), and their housemate came home, saw what was happening, and wisely sussed up the situation that there was some kind of deal going down, that she wanted in on.
So she put her own tip jar down at the end of her housemate's row of tip jars, and I blithely deposited $5000 into her tip jar several times, which she immediately snapped up.
When I realized what happened, instead of contracting The Sims Mafia [2] to do a hit on her, I congratulated her for her loose morals and ingenuity. It was such a great hack, and I totally fell for it, and had more Simoleons than I knew what to do with anyway. It's all about good customer service!
It was a fun experiment, but other bots and offshore farmers were starting to work the system too, and customer service and delivery problems made it not worth continuing.
So the unemployed Sims bot wouldn't feel bored, I retrained it into a more practical assistive utility called "Simplifier" [1], which knew how to recognize and navigate the Sims user interface to show, scroll through, and enumerate all the many items, wallpapers, floor tiles, etc, in the catalog.
It took snapshots of the icons, and read the text off the screen to capture the title, price and description (it was all in a bitmap Comic Sans font, so it was easy for a bot to recognize, if not for your eyes to read), and made a searchable illustrated database of all your built-in and downloaded content.
Simplifier addressed the problem that many players would download thousands of objects from web sites, or make their own custom objects with tools like Transmogrifier and RugOMatic (shown earlier in the demo video), but it was impossible to search or keep track of them through The Sims interface.
And it was useful for Sims web site publishers to make illustrated catalogs of their own objects.
You could also operate it in manual mode, where you press and hold on an icon in the catalog, and it reads the object description to you with a speech synthesizer.
That was useful for kids learning to read, old farts with bad eyesight, and snobby designers who hate Comic Sans, who would enjoy having the object descriptions read to them.
[1] Simplifier demo starts at 3:15: https://youtu.be/Imu1v3GecB8?t=3m15s
[2] The Sims Mafia: https://www.schneier.com/blog/archives/2009/11/virtual_mafia... and http://buildingreputation.com/writings/2009/10/the_dollhouse...
Napster -> Spotify
Chatbots (1996) -> Chatbots (2016)
Just because an idea failed, doesn't mean that it's resurrection is necessarily derivative or vacuous. Timing is everything, 20 years is a lifetime for applications of tech concepts.
[1] http://www.recode.net/2016/3/11/11586878/instacart-a-startup...
Idealabs approach me at one point offering to buy my botsinc.com dom.
Two things killed development. The first was the quality of the speech recognition. 85% accuracy sounds impressive, but in real life, it's terrible, especially combined with background noise. (I had a love bird that would squawk every time i tried to train the recognizer :) The SR has only recently caught up enough to be acceptable (just). The other was the AI. Every 'rule' had to be hand coded. eg: "if SR = 'hello world' then tts 'hello'". I don't think they'll get past that last issue anytime soon, and I believe the Loebner prize is still very safe.