I honestly wouldn't mind trying to work with ATM to better convey his AI 
approach than his current methods.

For instance, in one of his "earlier" journels, he references a "boulematic 
accumulator" - in normal lingo, it means neuron, like a neural network neuron.

So, things like that. Maybe he'd gain more credibility with better 
explanations. And... Less explanations.

Sent from ProtonMail mobile

-------- Original Message --------
On Feb 16, 2019, 7:52 PM, Steve Richfield wrote:

> Matt and Stefan,
>
> First, let's discuss the nature of reality. Our world is INCREDIBLY complex, 
> e.g. cells communicate with other cells via ~100 different channels. No 
> theory can accurately deal with this, e.g. Newton didn't consider Einstein, 
> Einstein didn't consider dark matter, and I suspect understanding dark matter 
> will still leave an imperfect understanding.
>
> ALL theories are wrong!!! Sure, there might be a correct one somewhere, but 
> if we ever run onto it, we will NOT be able to either recognize it or prove 
> its perfection.
>
> The Scientific Method provides a mechanism to prefer some imperfect theories 
> over other even more imperfect theories.
>
> Not 1% of 1% of "scientific" papers exist within the above reality, so they 
> are uncodable in a system that respects the basics of reality.
>
> Getting down to real world utility, I have developed new cures for several 
> illnesses, using the information contained in thousands of articles to form 
> theories to get onto a scientific track. I estimate that only ~1% of articles 
> contain ANY information that contributes to this process, and even that 
> information often emerges from between the lines.
>
> MATT: I seriously question the practicality or value of ANY system operating 
> on <1% S/N ratio, especially when there is often broad correlation in the 
> noise - which of course is why they have failed to find solutions. The most 
> obvious example came in eliminating my own glaucoma blind spots. It dawned on 
> me that everyone (else) might have simply presumed the wrong direction of 
> cause and effect - that vascular deterioration was impairing retinal 
> operation, rather than the reverse. Reversing this presumption quickly led to 
> a permanent cure using my own modest abilities, and my blind spots are niw 
> GONE.
>
> MATT and STEFAN: Facts have referrants. It is (nearly?) impossible to 
> accurately state ANY real world fact without a LOT of qualification - SO much 
> qualification that properly qualified facts, if they even exist, would be 
> worthless because they would only describe a single point in 4-D space-time.
>
> To get past this, people encode believed relationships, which by their nature 
> incorporate theories presumed to be accurate, when we know (or should know) 
> they are NOT accurate. "Modern" AI systems utilize knowledge bases in which 
> to encode these relationships, e.g. in an algebraic form, as I implemented in 
> my DrEliza proof of concept. DrEliza only incorporates ~200 such 
> relationships - close to what Matt calls "rules" but with some substantial 
> enhancements. Forcing expression into a particular algebraic form brings some 
> substational advantages, e.g. Bayesian math, Zipf presumptions to fill in 
> some missing data, etc.
>
> Legacy approaches (like Arthur has apparently utilized) code "rules" in a 
> procedural language, in what is often called "ad hoc" (unstructured) 
> programming. This has an advantage of raw speed to evaluate rules, and you 
> can hire programmers off the street and put them right to work, but misses 
> lots of important nuances like the ability to compute confidence, which is 
> REALLY important in high-value applications like medical diagnosis. Chatbots 
> (like Arthur has apparently programmed) are typically ad hoc programmed, 
> because speed is more important than precision.
>
> EVERY approach to AI is the BEST for SOME applications, so I find ALL of the 
> openly elitist postings on this forum to be rather irksome.
>
> What bothers me about Arthur is his unaddressed inability to state what he is 
> doing in terms that others here can relate to, and his inability to clearly 
> explain his dream based on this.
>
> Steve
>
> On Feb 16, 2019 3:00 PM, "Matt Mahoney" <[email protected]> wrote:
>
>> Steve, how many rules are in Dr. Eliza?
>>
>> How long did it take you to develop Dr. Eliza?
>>
>> How many rules would it take to encode the knowledge in all the world's 
>> published scientific papers (75 million)?
>>
>> On Sat, Feb 16, 2019, 1:51 PM Steve Richfield <[email protected] 
>> wrote:
>>
>>> Arthur,
>>>
>>> I have been one of your few supporters, but if you are going to usefully 
>>> engage with the present audience, you REALLY need to answer two questions, 
>>> that if done well will lead to other questions, that will lead to a useful 
>>> conversation...
>>>
>>> 1.  THEORY: In broad computer science terms, how does your system work? 
>>> From what I can tell, it is an ad hoc text manipulation program capable of 
>>> gathering information and answering simple questions within the limited 
>>> subject domains that have been programmed. Right?
>>>
>>> 2.  APPLICATION: What will your approach be able to do that the machine 
>>> learning approaches discussed here can never ever be extended to do, and 
>>> why? For example, my system works to diagnose chronic illnesses in a way 
>>> that can NEVER EVER be equalled with ML approaches. From what I can tell, 
>>> your system might be extended to make a really good military inventory 
>>> program.
>>>
>>> As with all AI programs, their authors have dreams for them that exceed 
>>> everyone else's expectations, and you and I are no exceptions. I understand 
>>> that ONLY ad hoc logic will EVER be capable of incorporating human 
>>> understanding of our world into a computer, a simple fact that is 
>>> universally rejected by others on this forum for NO good non-religious 
>>> reason. So, until others here wake up, at minimum, I should be able to 
>>> relate to your postings. If you can't carry me along, then you truly are 
>>> COMPLETELY wasting your time and your life by continuing to post.
>>>
>>> Steve Richfield
>
> [Artificial General Intelligence List](https://agi.topicbox.com/latest) / AGI 
> / see [discussions](https://agi.topicbox.com/groups/agi) + 
> [participants](https://agi.topicbox.com/groups/agi/members) + [delivery 
> options](https://agi.topicbox.com/groups/agi/subscription) 
> [Permalink](https://agi.topicbox.com/groups/agi/Tc360a468d4050822-M6fe5554cdf4507ca9bcd3ed7)
------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/Tc360a468d4050822-Mca56cdaf2e08d23f937de30d
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to