Re: cogAI vs AI KR characterization

Thank you Dave
I ll try to jot some of the concepts you mention on that slide
Human intelligence/cognition is such a vast topic, its useful to try to
narrow down the epistemic perspective
when designing AI however, transposing the cognitive model into explicit KR
is necessary, with some expectation perhaps
of some types of  genetic algorithms they call them ML/NN these days  and
which are not there yet, such as than a future model of the baby AI capable
of self learning
I d like to note a couple of points before the brain shuts down for the day
and resets onto other tasks
I work as a teacher, mostly for adults but increasingly enjoy teaching
kids, its a totally different experience which
has started feeding  materials into the research angle for cognitive systems
It must be noted that human learning/intelligence is very plastic and
context dependent
Physical conditions, such as nutrition, sleep  social conditions, such as a
stressful social environment such as family tensions
environmental conditions such as air and noise pollution, as well as the
availabilty of clothes, suitable desk and chair,
equipment, stationary, they all influence learning.   One of my students
used to sleep in my class, and  failed most
subjects consistently, btu she was nice, not a lazy person at all,  Only
after recommending a blood test it turned out she had sugar and iron
deficiency, some supplements and a change of diet fixed and the next year
she got B+ without trying much
So measuring human intelligence based on exams results is a gross
simplification

some subjects like maths or physics can be taught relatively easy, even
calcultators or a program can be written to
give the right answer  , not much thinking required, just apply the logic
in the right sequence and the result is correct
by default

however some subjects like essay writing or rethoric..... has anyone seen
an AI capable of rethoric?

Now we have codified the solution for a rubik cube, but before such
solutions existed, solving a rubik cube
took considerable more time and effort,  now even a robot can solve the
rubik cube
https://www.theverge.com/2019/10/15/20914575/openai-dactyl-robotic-hand-rubiks-cube-one-handed-solve-dexterity-ai#:~:text=In%202016%2C%20semiconductor%20maker%20Infineon,in%20less%20than%200.4%20seconds
.

Big topic

I ll expand on the slide when I can

p

but




On Sat, Feb 6, 2021 at 10:09 PM Dave Raggett <dsr@w3.org> wrote:

> See inline ...
>
> On 5 Feb 2021, at 21:08, Paola Di Maio <paoladimaio10@gmail.com> wrote:
>
> Hi Dave
> I am cc ing the lists, because this exchange is part of the discourse in
> our respective CGs and relates
> to a post to the lists
>
> well, I do study the child mind, but not with CogAI
>
> Because it is not yet clear what CogAI does in relation to other approaches
> (the slide aims to  help clarify)
>
> so, is the method of COGAI (as your emails suggest) mimicking? is there a
> reference for that?
>
>
> The idea is to match or improve upon human performance through executable
> software implementations of functional models inspired by observations and
> theories of human behaviour, especially that of children.
>
>
> you write
>
> * If we can successfully reproduce how the best people reason,....*
>
> how does COGAI defin best people ?
>
>
> That would depend on what you’re looking for. One metric is how good
> people are at passing exams.
>
> Human-like AI is perhaps a better term than Cognitive AI as it makes it
> immediately clear that the focus is on mimicking human abilities.
>
>
>
>> an afterthought
>>
>> in respect to mimicking how humans reason and communicate well,
>> each human  is different, we can generalize up to a point
>>
>> and mimicking may result in some kind of parrot engineering ....
>> useful to start with but nowhere near intelligence at its best
>>
>>
>> You’re missing the big picture.  If we can successfully reproduce how the
>> best people reason, we will be in a strong position to improve on that by
>> going beyond the limits of the human brain. The more we understand, the
>> further and faster we can go. This is an evolutionary path that will go
>> very much faster than biological evolution. At the same time we can make AI
>> safe by ensuring that it is transparent, collaborative and embodies the
>> best of human values.
>>
>> Human-like AI will succeed where logic based approaches have struggled.
>> 500 million years of evolution is not to be dismissed so easily.
>>
>> I remember the enthusiastic claims around “5th generation computer
>> systems” and logic programming at the start of the 1980’s, and had plenty
>> of fun with the prolog language. However, the promise of logic programming
>> fizzled out. Today, 40 years on, much of the focus of work on knowledge
>> representation is still closely coupled to the mathematical model of logic,
>> and this is holding us all back. We need to step away and exploit the
>> progress in the cognitive sciences.
>>
>> I am especially impressed by how young children effortlessly learn
>> language, given the complexity of language, and the difficulties that adult
>> learners face when learning second languages. Another amazing opportunity
>> is to understand how some children are so much better than others when it
>> comes to demanding subjects like science and mathematics. Moreover, warm
>> empathic AI will depend on understanding how children acquire social skills.
>>
>> Let’s lift up our eyes to the big picture for human-like AI.
>>
>> Dave Raggett <dsr@w3.org> http://www.w3.org/People/Raggett
>> W3C Data Activity Lead & W3C champion for the Web of things
>>
>>
>>
>>
>>
> Dave Raggett <dsr@w3.org> http://www.w3.org/People/Raggett
> W3C Data Activity Lead & W3C champion for the Web of things
>
>
>
>
>