Creating Information 4.0
Fei Min Lorente
mba...@analecta.com
I think that conversation took a left turn at Albuquerque. Information 4.0 seems to be a piece of argument by analogy that suggests that what comes next is self-assembling chunks of nano-content: http://www.informationenergy.org/news-articles/detail/information-40-the-next-steps/.
But I think that the real promise is in Information 5.0. Arguing by analogy again, and by extension (as we must, since all trends continue), if information 4.0 does away with those lumbering great “topics” of Information 3.0, breaking content down into far more granular phrases and words animated by intelligent bots, Information 5.0 will provide the ultimate sophistication by breaking everything down into a mere 26 sub-word components manipulated by intelligent agents capable of independent learning and creative thought. In fact, information 5.0 already as a catchy theme song:
https://www.youtube.com/watch?v=a9UJSIGWl7Y
Now, if I were to be just slightly less snarky than usual, I would say that I think that the point buried in the hype might be this: if we build machines that can learn, how do we get them to tell us what they learned? This is not trivial, because we have yet to figure out how to get humans to tell us what they learned. Tell me how to catch a ball. Tell me how to walk. We learn these things but we have no words for them. We have invented entire science and engineering disciplines to try to figure out how we do these things and to try to reproduce them in machines, and they still can’t come close to the mobility or agility of a my four year old granddaughter.
A related problem is, if the machine learns to do something, how do we tell people who buy, license, and regulate the machine what it is going to do. Most machines today, at least nominally, are supposed to come with documentation that tells us how they are going to behave given a certain range of inputs. That is harder to do if the machine learns from each set of inputs it gets. But then again, most of our machinery today is so complicated that we can’t actually tell you this now. Merely multiplying permutation is enough to make the range of possible outcomes practically unknowable. (There were some on the Manhatten project who feared that setting off an atomic bomb would start a chain reaction that would incinerate the planet. And someone must have said words to the effect of, “Okay, but let’s do it anyway and see.”)
One of the points I tried to make at the dinner was that there is nothing new about machines that make decisions over which we have no understanding or control. The traction control system in your car is an example. The amount of data gathering and control inputs that go into keeping you out of a ditch are staggering, and every skid is unique. A machine took in information and applied control inputs to a vehicle travelling at speed, overriding driver inputs, in the middle of a busy roadway filled with other cars and lined with pedestrians. This is not only legal, it is mandated by law, and no one can tell you completely and certainly, what it is going to do in every situation it encounters.
The difference with AI, therefore, is not that we are suddenly in the dark whereas once we were in the light, or that we are not in control where once we were in control. We have have been careening in the darkness for a while (though, curiously, not crashing into things quite as much as we used to). The point with AI is not that is to going to be something other than algorithms processing data, which is exactly what it is now, it is simply that the algorithms are going to be developed differently, which is to say that algorithms are going to be written by algorithms which take information from one interaction with the world and reject decisions that result in bad outcomes and replace them with new decisions until they find one that produces a better outcome.
That is pretty much what developers do now, of course: write different algorithms, throw away the ones that don’t work, keep the ones that do, then ship when enough tests pass. With AI the machine does the same thing, except it keeps doing it in the field, and, potentially, it shares what it knows with other machines in the field, so your traction control system keeps getting better and better until it sets a new Olympic ice dance record.
This will mean that we won’t know how our traction control system algorithm works on any given Thursday (this was a topic of some heated contention over dinner). The algorithm was written by the machine and its buddies who swapped improvements over the IOT the way developers swap coding tips on Stack Overflow today. But the truth is, there is rarely one person who can tell you how todays algorithms work in their entirety either. Very few are the products of a single mind, and fewer still are adequately documented, or documented at all, or even capable of being documented comprehensively given the possible permutations of their actions.
We could, theoretically at least, throw our current algorithms on a test bench and do an exhaustive analysis and completely understand everything they did, though there is the small problem that no one mind could contain the whole of some of them, nor live long enough to complete the analysis. Algorithms could help, but, well, you see the problem with that…
In short, we don’t completely know, and cannot completely regulate, how our tools work now, and after AI comes along, we still won’t, only more so.
But take heart, because this is actually not a new problem. In fact, since the dawn of time we have use a tool that learns by rewiring its algorithms in response to experience, and whose algorithms we have never fully comprehended. It has helped up gather food, fight wars, guard our homes, businesses, and cities, raise our children, travel through inhospitable terrain, guide us, and comfort us. Despite all these virtues, it is still widely feared, sometimes bites the hand that feeds it, and occasionally tears the throats out of your neighbor’s children.
Here it is:
It is still in the development stages, but we are hoping that Dog 4.0 will know how to tell us what it learns, what its exact capabilities are, and how it will react in any given situation. The technical term for this is a “chatmutt”.
Mark
--
--
You received this message because you are subscribed to the Google Groups "STC Southwestern Ontario".
To post to this group, send email to stc-southwes...@googlegroups.com
To unsubscribe from this group, send email to stc-southwestern-o...@googlegroups.com
To change your personal settings, visit this group at http://groups.google.com/group/stc-southwestern-ontario?hl=en?hl=en
Privacy Statement
The STC Southwestern Ontario Chapter (STC-SOC) will not sell, rent, or give away your email address to any outside party. This list is strictly for discussing technical communication issues and announcing technical communication news and events. If you do not wish to receive STC-SOC emails, please send an email to stc-southwestern-o...@googlegroups.com.
---
You received this message because you are subscribed to the Google Groups "STC Southwestern Ontario" group.
To unsubscribe from this group and stop receiving emails from it, send an email to stc-southwestern-o...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
Fei Min Lorente
To post to this group, send email to stc-southwestern-ontario@googlegroups.com
To unsubscribe from this group, send email to stc-southwestern-ontario+unsub...@googlegroups.com
To change your personal settings, visit this group at http://groups.google.com/group/stc-southwestern-ontario?hl=en?hl=en
Privacy Statement
The STC Southwestern Ontario Chapter (STC-SOC) will not sell, rent, or give away your email address to any outside party. This list is strictly for discussing technical communication issues and announcing technical communication news and events. If you do not wish to receive STC-SOC emails, please send an email to stc-southwestern-ontario+unsub...@googlegroups.com.
---
You received this message because you are subscribed to the Google Groups "STC Southwestern Ontario" group.
To unsubscribe from this group and stop receiving emails from it, send an email to stc-southwestern-ontario+unsub...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
--
--
You received this message because you are subscribed to the Google Groups "STC Southwestern Ontario".
To post to this group, send email to stc-southwestern-ontario@googlegroups.com
To unsubscribe from this group, send email to stc-southwestern-ontario+unsub...@googlegroups.com
To change your personal settings, visit this group at http://groups.google.com/group/stc-southwestern-ontario?hl=en?hl=en
Privacy Statement
The STC Southwestern Ontario Chapter (STC-SOC) will not sell, rent, or give away your email address to any outside party. This list is strictly for discussing technical communication issues and announcing technical communication news and events. If you do not wish to receive STC-SOC emails, please send an email to stc-southwestern-ontario+unsub...@googlegroups.com.
---
You received this message because you are subscribed to the Google Groups "STC Southwestern Ontario" group.
To unsubscribe from this group and stop receiving emails from it, send an email to stc-southwestern-ontario+unsub...@googlegroups.com.
mba...@analecta.com
Well, if it means the experience quietly upgrading itself in the background over time, such as your traction control getting better and better the more all the cars in the world recover from skids, then I’m not sure why the docs have to say anything other than “Gets better every day!”. And let’s not forget that most of our apps and many of the desktop programs we use silently upgrade themselves in the background now. The agency may change, but the effect is already with us.
Are AIs going to evolve new features and new interfaces by themselves? If they do, they will manage and roll them out the same way the human developers do now. They will maintain a separate development branch of their code, test iterations on beta users, document new editions, if required, and then roll them out into general release after a period of testing. Developers create and run thousands of versions of an app, most of which are never run outside the development organization. From there maybe a hundred go to test, and maybe 10 go to beta, before one goes live to the public. There is no reason for that to change just because robots are writing the code.
One of the limitations of the current development model is that security and performance upgrades go out through the same mechanism as feature and interface upgrades, even thought these upgrades have different natural schedules and different kinds of consumer impact. The web has allowed development organizations to separate these things to a useful degree. AI and the IOT may allow them to further separate them, further increasing security and performance upgrades while making it much easier to test and to roll out feature and interface upgrades on a schedule that works best for humans.
Of course, the whole chat bot thing is an attempt to do away with interfaces altogether. Intelligent chat bots will make user interfaces unnecessary, just as intuitive user interfaces made documentation unnecessary. (They did that, right?...)
Dog 1.0 has a much reduced role in our lives lately, but that is just the last eye blink of human history. Your hunting dogs, which you have trained to run down and kill small animals, sleep by the fire with your children. They are your security system, one of your most important appliances, and possibly your transportation system at all.
Dog. Horse. Cave. Fire. Knife. Stick. Everything else is just a version upgrade.
Mark
From: stc-southwes...@googlegroups.com [mailto:stc-southwes...@googlegroups.com] On Behalf Of Fei Min Lorente
Sent: Sunday, December 3, 2017 10:50 PM
To: stc-southwes...@googlegroups.com
Subject: Re: [STC-SOC:353] Creating Information 4.0