Can’t the Big Players just do what you’re doing?

Well, yes, in principle.

Are they likely to? No. It isn’t in their DNA.

One could also ask: Why couldn't Microsoft with all of their billions stop upstart Google? Or Google with their resources decimate early Facebook? Or Walmart compete with little Amazon?

Large company are incredibly good at leveraging their core strengths. The particular strengths related to AI are huge amounts of data and lots of processing power. That’s their hammer, and every business problem or opportunity looks like a nail. Their strength, focus,and expertise is in Big Data and Machine Learning — essentially statistical approaches.

Statistical AI focuses on quantity, not quality. Sophisticated conversational comprehension on the other hand requires maximal quality. A 2017 DARPA report stated, ‘Current AI methods are statistically impressive but individually unreliable.’

Advanced, adaptive, natural language assistants cannot be careted with statistical methods alone, they require some sort of cognitive architecture, also referred to as ‘The Third Wave of AI’. Current chat-bots and so-called ‘personal assistants’ are all implemented using first and second wave AI technology.

There’s almost a ‘perfect storm’ for companies single-mindedly focusing on ‘second wave’, statistical methods: Over the past few years these methods have become incredibly successful in many applications areas such as speech and image recognition, translation, and categorization. Because of this it is significantly easier to get a PhD sponsor, media attention, startup funding, earn top dollar, or be acquired by pursuing statistical AI such as deep learning/ machine learning.

Furthermore, most software engineers and data scientists are much more comfortable thinking like mathematicians or logicians rather than cognitive psychologists — something that is essential for designing successful cognitive architectures.

Additionally, large companies, like oil tankers, change course slowly. They have strong incentives to incrementally improve existing systems rather than making radical changes:

• to meet short-term financial objectives
• to leverage existing knowledge and infrastructure (the legacy trap)
• to minimize risk by sticking with standard technology

These factors conspire to ‘suck the air’ out of alternative AI approaches such as the cognitive architecture required to achieve real conversational intelligence.

“Electric light bulbs did not come about from the continuous improvement of the candle …“ — Oren Harari

PS. “How do you know that there isn’t some secret cognitive architecture R&D project at GoogleX or similar?”
Of course, we can’t know for sure, but the cognitive architecture community is actually quite small and one would tend to notice if several of them ‘disappeared’. We haven’t seen that. Secondly, even if there were such projects (a) what we’ve developed is actually really hard to replicate (and we all know that 100 programmers often can’t achieve what the right team of 5 can); and (b) they would still have to be supported by top management to get into full production. Here they would face many of the hurdles outlines above. A good example of this is Google’s abandoned ‘Orkut’ service — a very good Facebook-like site before FB really took off. Google management didn’t give it support it needed and was thus deathly slow. That, among other issues killed it.