How Does a Neural Network Work?
A neural net is not much more than a collection of logical units that each has an input weight and a weight per connector. If the input weight multiplied by the connector weight is more than a threshold value defined, the unit is set to fire, which then triggers the unit to its right with a new input value.
In the graph on the right, due to the fact that not all threshold values are exceeded, the tree will shrink until only a couple of output units or even only one have fired. This is the basis of how it learns.
To illustrate the fact that actual learning took place, consider the AlphaGo AI program that was able to understand the goals of one of the most difficult games ever created, and a result of a learning process, some out of the box “thinking” assisted it to beat the human champion.
But hold on, you say: a board game is still a board game and that could map easily to how the neural net (see diagram above) works, but does something based on simple logical gates above really take us closer to General AI and language understanding?
After all, language has always been a bastion of humanity because language is so difficult to master. That said, Open AI has changed all this by finally developing an AI that understands language. This AI is known as GPT-3.
It understands language in a multitude of variations, and it can process complex statements or subtle hints. It can even feign ignorance.
If you input a paragraph of text, it can output a paragraph of text that finishes your thoughts. If you ask it to summarize a book, it can offer a comprehensive summary. It can assume the tone of various writers and even translate common commands into complex programming code.
Consequently, all the jobs based on language are now susceptible to being performed by AI. This article you are reading is being written by someone who might very soon be out of a job. Have sympathy, because it is not just writers that will be impacted. If your job involves language, you, too, could be next.
Web design requires a designer to understand three things: coding, graphic design, and a customer’s desired functionality. Simply put, it requires a mix of technological muscle and intuitive art.
However, GPT-3 can accept a sentence or two about what a person needs in a website and create a web page within seconds. The coding is in place. The text contains all the CSS customizations. The graphic design can be based on a similar site. The end result is an attractive, functioning website generated within a few seconds.
Here’s a sentence describing what Google’s home page should look and here’s GPT-3 generating the code for it nearly perfectly. pic.twitter.com/m49hoKiEpR
— Sharif Shameem (@sharifshameem) July 15, 2020
To create something that takes on the abilities of other professionals is one thing. To create something that learns and masters your own abilities hearkens of self-sabotage. That said, in the same way GPT-3 can put together complicated web pages, it can also program apps for your phone. All that is required is for the person who wants the app to describe the functionality. GPT-3 then takes that description and programs the app in such a way that the app now functions as desired.
This is mind blowing.
With GPT-3, I built a layout generator where you just describe any layout you want, and it generates the JSX code for you.
W H A T pic.twitter.com/w8JkrZO4lk
— Sharif Shameem (@sharifshameem) July 13, 2020
This type of programming is achieved by GPT-3’s ability to understand language, translate that language into pseudo-code behind the scenes, and generate functional code. It knows when to insert conditional functions, and it can infer from a user’s words what is required to complete complicated programming.
Artificial intelligence is already disrupting the legal field, but GPT-3 is doing something few other programs can do.
First, it understands the intricacies of legal jargon and can translate it into so-called everyday language for anyone to understand.
Just taught GPT-3 how to turn legalese into simple plain English. All I gave it were 2 examples 🤯 Might build a term sheet and investment document interpreter out of this 🤓 pic.twitter.com/BDdwCuFce5
— Michael (@michaeltefula) July 21, 2020
Second, in the same way it can diagnose medical problems based on evidence, it will soon be able to advise on legal strategies based on details of a case.
Finally, it can prepare a variety of customized legal forms on the fly.
As such, it will soon be able to take on the task of preparing documents. Because it can summarize in regular language, it will be able to serve as a legal interpreter. Finally, once upper-level partners become convinced of its powers, the lower-level junior attorneys may no longer be needed.
For SEO writers, GPT-3’s ability to write content and match all the necessary SEO parameters to scale Google’s search ranking is an easy task. Furthermore, its ability has already been shown by a college student who used a demo version of it to write fake blog posts that ended up headlining Hacker News.
Being that the expertise to create quality SEO already exists, all it will take is for a subscription service to be implemented in which content publishers can subscribe to a content service and receive computer-generated content.
GPT-3’s ability to understand language will not impact writers immediately, but within 12 to 14 months, it is probably that a complete novel will have been written using nothing but artificial intelligence. This does not bode well for writers because GPT-3 will likely be able to soon discern what types of language have the best effect and what types of plots generate the greatest interest. If you combine these two functionalities with being able to then customize a novel to a particular reader, you have just outlined the perfect nightmare for every professional novelist.
Also Read: [7 Impacts of AI on Our Future]
This scenario is not far-fetched. For instance, GPT-3 can already take a paragraph of narrative and then extrapolate events, characters, and plots. The demo version can do this with one paragraph, but Open AI will no doubt develop a version that can do this on a grand scale of 100,000 or more words.
Finally, because GPT-3 will be able to understand what a reader wants, it will be able to generate a custom novel for anyone. As it can already generate images of people as well as art, it will easily take over the jobs of graphic artists who specialize in graphic novels.
GPT-3’s ability to understand language and output graphics as well as code make it poised to begin competing with game developers. However, this will take a little more time as it depends on a variety of complex coding examples that GPT-3 has not been shown to have mastered.
Although GPT-3 will not be prescribing medication any time soon, it can already accurately assess a variety of symptoms and come up with an accurate diagnosis. As such, it will likely become a valuable consultative aid for physicians. In the long-run, it will replace diagnostics assistants, helping to screen patients and make triage more efficient.
Still don’t believe the cognitive level of GPT-3? See the interview below.
GPT3 represents the end of the line for workers in general in that it has finally accomplished full understanding of natural language. The only thing slowing its disruption of multiple industries is that corporate owners and Open AI founders are attempting to prioritize the ways in which it can be implemented.