And another reason we don’t have AGI yet

My Iraqi friend (whether he knows it or not) was on TV again tonight. God bless the man (his god or mine or whatever) because i always get an idea or two out of his broadcasts. For the record, his name is Jim Al-Khalili, and this time the show was called Shock and Awe: The History of Electricity. I can’t help but throw the man some props for his choice of show title, him being from Iraq and all, although i’m not sure if he attended the second run of Desert Storm or not.

Anyway, lately i’ve been playing around a bit more with the LIDA framework, which i’ve described in my last post. After several discussions with Ryan McCall (PhD student at U of Memphis and apparently the current primary LIDA maintainer), i’m still very hopeful about the potential of the work. But i do now, i think, have a better handle on where it stands. The framework as it is represents a good implementation of a good theory of cognition. It has at least interfaces for all of the major components, and in many cases concrete implementations of those components. It is an excellent start. Of course, there are components that are really just stubs at the moment, and other components that are very basic, and they will need work before any serious run at AGI (with LIDA) can be made. And even when such a run is made, many of the more complete components will need substantial rework, and it’s pretty much certain that the overall design of the framework will go through many iterations (which, each time, will have major implications on all of the existing components). All of which brought me to the conclusion that there is a lot of work to do. A lot. Like years. Decades even. Maybe lots of decades.

Kind of like electricity. I know that the typical AGI comparison is with human flight, but after watching only a bit of Jim’s show i wonder if electricity is a better one. Flight had all of it naysayers in the face of existential proof and all, and people wondering what the point of it all could be anyway, but after only a bit of research i think the study of electricity provides an analogy that parallels the profound scientific mysteries of the subject.

According to Wikipedia, the earliest recorded recognition of something like electricity was from ancient Egyptian texts that described the “Thunderers of the Nile” – now known as electric fish – back in 2750 BC. The Egyptians either killed all the little buggers off or shrugged their shoulders and stayed out of the water, because there doesn’t seem to be much that happened on the topic for around 4.5 millennia, when William Gilbert in 1600 described the difference between the loadstone effect and rubbing a balloon on his head. If we take this as the real start of electrical research, we need only decide on when we feel we had a decent handle on the topic to determine how long the whole effort took. Maxwell’s work in 1861/62 seems like a reasonable choice except that it was only theoretical. Again, Wikipedia provides great satisfaction to me if only by its choice of words: “… the late 19th century would see the greatest progress in electrical engineering” (my emphasis).

We can then conclude that it took about 300 years to convert electrical study from research into engineering. Readers of my last post will recall that i made a big deal about the difference between treating AGI as research vs engineering. When i first started working with LIDA i had hoped that we were entering the engineering stage, but it appears my hope was premature. There is still a lot of research to do. Those who are frustrated that after over 60 years we still have little to show might take some comfort in knowing that something similarly mysterious, but relatively simple, as electricity took nearly 3 centuries for some very big-headed fellows to get their heads around.

Ray Kurzweil was most likely optimistic in predicting the existence of AGI by 2035. Which sucks, really, because i was very much looking forward to climbing into my Vanilla Sky reality simulation over being spoon fed by R2D2 prototypes in a seniors’ home.

To round off… The other reason we don’t have AGI yet? It’s really, really hard.

10 thoughts on “And another reason we don’t have AGI yet

  1. Correction, U of Memphis, but I’d rather be associated with TN than Memphis, lol.

    I generally agree with your assessment about LIDA and AGI, however, it’s not clear that all the separate modules of LIDA will be *so* different in the future. What I think will happen is that we’ll move from a “kitchen sink” to an architecture with more uniformity underlying the “modules” (which are often imaginary boundaries). I think that by incorporating a common cortical algorithm a la Mountcastle & Hawkins as well as ideas falling out of Friston’s free energy principle we will find more elegant and effective ways of accounting for the whole of cognition is a relatively simple way. =)


  2. Doh! Sorry, i’ve corrected the post. Of course I could say you’re at Harvard or something if that’s even better.

    I used to like the “cognitive algorithm” idea, but over time i’ve become more skeptical. I know that cortex is very flexible and can be used or reused by different brain structures in cases of damage or extremely specialized learning (i.e. the size of a violinist’s homunculus’ hand), but to me this suggests, rather than being the seat of intelligence, that it is a relatively simple peripheral, like a the-bigger-the-better hard drive. (I hate making analogies to computer hardware, but there it is.) Under cortex there are probably hundreds of specialized little brain areas that are all wired differently, do different things, and probably use cortex to support their operation. As you say, lines between the real “brain modules” are blurry, and LIDA modules are no doubt idealized, but i’m not hopeful that there is a relatively simple way of explaining all brain activity at a level beyond neuron firing.


  3. Matthew, Just stumbled across this site, very nice read. Regarding the argument of single cortical algorithm (like Numenta’s HTM) versus “kitchen sink” (like LIDA) I wonder where all these “hundreds of specialised brain areas” are stored on the human genome. Human DNA is only 50 MB and contains descriptions of other organs and minute cellular details. There’s also lots of junk DNA. I just can’t imagine how the algorithm for these “hundreds of brain areas” can be stuffed into such a small amount of storage. And there is also the issue of how the “kitchen sink” could have evolved. I think if the human brain turns out to be like LIDA, creationists will be delighted! Since I think creationists are wrong I suspect the human brain is nothing like LIDA.

    It says in New Scientist issue 2806 that “LIDA is a high level model of the mind and does not attempt specific neurons or brain structures”.


  4. Thanks for reading, John, and for your comments.

    LIDA is based upon Baars’ Global Workspace Theory. Neither LIDA nor Baars draw parallels between theoretical components and brain structures; that’s all me. Still, it is a theory of mind, and if it is anything like accurate i think it’s reasonable to expect that such parallels exist.

    I don’t know how different brain structures form, but then i don’t know how hearts, livers, lungs, pancreas, hair, 206 different bones, fingernails and feet form either. But it is undeniable that they indeed do. The hippocampus, cerebellum, thalamus, pons, and corpus callosum et al are all distinct from each other and cortex. Perhaps DNA information overlaps in very complex ways allowing it to store more than 50MB. Or maybe junk DNA is only being called that until we find out how cleverly utilized it really is. I’m just sayin’… Also, note that DNA is not all that determines a new human being. There is a great deal of contextual information carried in the cell that holds the DNA (such as mitochondria). I’m no biologist, but I suspect much of the cellular detail you mentioned is actually passed outside of DNA.

    Creationist might be happy to hear me say things like this, but their joy will be short-lived, i’m afraid. No matter how vastly complex the brain is, we will eventually figure it out, at least enough to recreate it in a different substrate.

    Ray Kurzweil would appear to be on your side though. I’ve been meaning to write an entry on his new book. I’ll try to get to that soon.


  5. Thanks for your reply, Matthew.

    I guess like Mulder I “want to believe” that there is a single cortical algorithm as it makes things much simpler and more elegant. It also makes recreating the human brain seem more feasible. But your right in saying there are many organs in the body (a “kitchen sink”) so given that there is no single “body algorithm” why should the brain have a single algorithm?

    We think of the brain as one organ but perhaps it is many. Perhaps there are more “organs” in the brain than the rest of the body combined!

    It’s a shame if there is no cortical algorithm as it means the work of Numenta and Vicarious is doomed from the start, and they seem to have so many good ideas and motivation. On the other hand there are projects like LIDA that come from the other end so, given researchers are covering so many bases perhaps one will finally get the catch. If LIDA turns out to be the way the human brain works I’m all for it and to hell with Occam’s razor!

    Speaking of LIDA, have you seen any demos of what “she” can do? Stan Franklin et al are making some impressive statements about LIDA’s capabilities leaving you to think AGI is all but complete with only minor arguments remaining about whether LIDA is functionally or phenomenoligically conscious. Since I can’t prove whether or human being are conscious I’m happy to let that slide.

    Yet with all these claims I’ve yet to see any trascripts of LIDA doing really intelligent stuff. Have you seen anything like that? Heck, why isn’t LIDA passing the Turing test yet? Surely she could at least win the Loebner prize!

    It’s great that the Memphis guys have put all these tutorial online but I’d like to see some demonstrations. Are any such things available? I’d be shouting fro the rooftops it I were them…


  6. Shouting from the rooftops might be a bit premature to my knowledge. I’m aware of the work that the LIDA group did with the U.S. Navy (from their description only) and the simple example that ships with the software. If you’ve seen evidence of other implementations, maybe you can point me toward them and i’ll check them out.


  7. The main one I’m thinking of is the US Navy. I’m not aware of any other implementations which is perhaps problematic since no transcripts are published to back up the claims being made. The claim is that IDA can negotiate with sailors sending and receiving unstructured email messages. This is extremely impressive, way beyond IBM Watson in my opinion as the latter can only answer one-sentence questions. And IDA’s natural language abilities seem to stem from the whole Global Workspace theory with various codelets combining in some way to give context and action selection. (i.e. the conversations are not simply pre-programmed). IDA can remember the context of previous conversation and continues on that basis.

    Furthermore, IDA can apparently feel emotions like frustration and nervousness while involved in these negotiations. That’s a strong claim! And of course there is the oft-repeated claim that (L)IDA is conscious. (though the definition of consciousness seems to be codelets broadcasting, not perhaps what many would expect).

    Stan Frankin reports “repeatedly watching Navy detailers nodding their heads and saying ‘Yes, that’s how I do it’ while watching IDA’s external and internal actions”. I’d love to see what sort of evidence made the detailers respond so positively. Just a few transcripts of emails would be enough to convince many. There barrier to learning IDA is quite high so enthusing people with examples would be nice.

    So there is plenty to shout from the rooftops and Franklin et al seem to be doing that with some jaw dropping claims. The trouble they are not backing this up with evidence as far as I can see. Of course it may be they are restricted by Navy policies on this. They have released the code but does it come with really compelling examples? I’m afraid there may be some “secret sauce” in the Navy (IDA) implementation.

    So what’s your take on this? Having installed LIDA and tried using it, do you think the examples included and your own trials are enough to convince you the claims are true. Is LIDA capable, in principle, of negotiating with humans and feeling emotion? Do you think LIDA is functionally conscious?


  8. I did install LIDA, and i worked with it for a while… spare time over the course of a few weeks. I even submitted some small enhancements.

    I have no basis for determining whether LIDA feels emotions. What ships in the download is the framework only. It’s up to the developer to implement the desired behaviour, and this is no small task. Also, as i mentioned before, some of the modules are basically stubbed out (as of my last checkout from the repo), so if you want, say, declarative memory, you’re on your own to actually build the module; the framework merely facilitates the passing of current context to it. The perceptual associative memory is implemented, but only as a slipnet. This may be sufficient – it’s hard to know – but it seems overly simplistic in my opinion.

    And the very nature of the framework, which i have praised as the right way to do things, makes it maddeningly difficult to analyze and debug what it going on. E.g. following a trial run you want to know why it didn’t do what you expected. Don’t expect to be able to set a breakpoint and step through the code until you say “ahah! There it is!”, and fix the bug. It’s more like trying to monitor wind currents from millions of weather stations trying to determine what breeze started the hurricane.

    If the framework is all that Franklin says, then, i agree with you, he will have to provide the evidence. Then, clever people will start using it and develop tools to simplify the creation of new and diverse agents, which will help build the LIDA community and get the ball rolling.


  9. I’ve installed LIDA and got the AlifeAgent demo working but haven’t had time to grapple with what’s really going on. I can see the program is enormously complex. I’ve listened to some of Stan Franklin’s lectures on the website but it’s quite philosophical — nothing that looks like an algorithm. I think some more detailed documentation is needed perhaps at the pseudocode level (not program language specific).

    It’d be nice to have something like a travel agent implementation in LIDA to show the capabilities they are claiming with the Navy version in open source. Do you think this is possible for 3rd party programmers?

    Anyway, I’d very much welcome a blog post describing your adventures with LIDA! (hint hint!)

    (also tried to register to GoID but couldn’t get in — sent you email asking for help…)


  10. I agree. What would be required is a very well documented practical implementation of something like, as you said, a travel agent. Something sufficiently complex that the capabilities of the software can really be seen. In order to educate myself, i created my own implementation, but it’s very simple (about as complex as the ALife agent) and pretty much undocumented. If you want to take a look though, here you go:


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s