Is AI Dumb?

Let's ask a robot.

February 3, 2020

Current AI systems are not intelligent because they were introduced - they are a memory. Unfortunately, the term "artificial intelligence" is becoming more common and it is more applicable to anything automated.

But what is true intelligence?

According to Wikipedia, "Intelligence ... is the ability to perceive information, or to manipulate it, as knowledge applied to positive behaviors in the environment or context."

From this point on, any service that changes its behavior due to new data is really smart. But it lacks the idea of ​​people's intelligence as self-awareness, learning and emotional cognition.

To define whether the system is good or not, the industry generally defines several KPAs of the systems (e.g. target completion rate, return rates) and low percentage (usually depending on the customer and their goals).

How should AI change/innovate to become really smart?

The main drawback of combining sample analysis is e.g. Machine learning (when you have gigabyte data) with specific rules (meaning domain or case-specific). Why? Classical machine learning methods find some rules and patterns in the data, but what if you changed the data? Re-train your neural network.

Perhaps the upcoming Integral Neural Chips (CPU / GPU for Performance Training of Neural Networks) will reduce the learning time, but does not solve the problem. Without simple AI, your service should be customized to work for you.

Is there a step before General AI?

General AI may appear in the next 10, 20 or 100 years - no one knows for sure when. This is the biggest challenge because, without a simple AI like a real human being, we wouldn't be able to get to the "advanced" level. Why? For example, a child does not need to bite the dog thousands of times to understand "it is doggo". Not so in the case of machine learning - “Train me with millions of examples. So, you practice, but do your photos of Doggo come from another angle? I didn't see any dog. "

This is why image recognition and chat and voice bots misinterpret you and your data. Have you implemented a great chatbot for an insurance company that solves 95% of consumer requests? It can only address 10% of consumer requests if you are connected to the banking industry. Computers are still fast enough to make computations, but they can't solve "meaningful" problems.

What is your approach? How do you do things differently?

So, in this case, use the formula I read in Peter Thiel's "Zero to One."

"Don't oppose the man, the machine, but work together."

What we do for our sound talk bots: The bot asks and answers the situation, but when we find out something is wrong - connect smoothly with a real human agent (you can't hear it).

According to our statistics, if people don't solve their problem with the voicebot they talk about in their first experience, they will refuse further conversations with 63% of potential bots. We call it "tandem help".

Best example - Excostleton for people against a full Android robot (which improves their chances).

In my opinion, according to past experience, I see it as "more than a fully automated world" (note the movement of the Luddites, who eventually became servants of such machines).


Leave a Reply

Your email address will not be published. Required fields are marked *

(c) 2020, a Codecide company.
linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram