Thursday 23 July 2020

Will The Latest AI Kill Coding?

This is the question that was asked in an article from towards data science on July 21st 2020. The answer follows: AI can now code in any language without additional training. The article goes on to say:
In 2017, researchers asked: Could AI write most code by 2040? OpenAI’s GPT-3, now in use by beta testers, can already code in any language. Machine-dominated coding is almost at our doorstep. GPT-3 was trained on hundreds of billions of words, or essentially the entire Internet, which is why it can code in CSS, JSX, Python, — you name it. 
The article continues:
Here’s the situation: Beta testers are using GPT-3 to generate working code, with trivial knowledge necessary. From buttons to data tables, or even re-creating Google’s homepage. These examples are all being done with zero-shot learning. Besides the rapid evolution of AI, two other major tech trends are compounding the reality that programming jobs won’t be safe in the future: No-code and AutoML.
No-code refers to visual tools that make it easier for anyone to build new products, whether it’s websites, designs, data analyses, or models. WordPress, Wix, and Shopify are good examples of no-code tools that enabled millions of people to do things on their own rather than hire a developer or a designer. The second trend is AutoML, or automated machine learning, which drastically shortens the time it takes to bring AI to production.
Tools like Apteo combine these trends, and enable anyone to deploy AI models with no coding skills required. GPT-3 will spark an additional wave of no-code and AutoML tools. Many would-be employers will opt for these tools rather than hire expensive programmers. Naturally, the lowest-skilled swaths of programmers will be the first to go, while experts will enjoy job security for longer — the same as in any field. 
 As the OpenAI website is offering free testing of its API and many more uses of GPT-3 will emerge as a result. According to this paper:
“GPT-3 achieves strong performance on many NLP (Natural Language Processing) datasets, including translation, question-answering, and cloze tasks, as well as several tasks that require on-the-fly reasoning or domain adaptation, such as unscrambling words, using a novel word in a sentence, or performing 3-digit arithmetic”. The team also finds that “ GPT-3 can construct samples of news articles which people will get conflict in order to recognise articles transcripted by humans.”
This sort of news only reinforces what limited job opportunities will be available to the youth of the future. It also means that, as with deep fakes, we won't be able to tell what is real anymore. Future versions of software tools created from GPT-3 type software will be able to generate a plausible story about anything and even populate it with deep fake images and videos. A bewildered public will be forced to rely on accredited news sources because the amount of fake news generated by this type of AI will be overwhelming. Of course, the "accredited news sources" will be free to propagate their own fake news as usual but it will be officially sanctioned and therefore trustworthy.

No comments:

Post a Comment