Artificial Intelligence Will Write Code Itself.
The future is already here. How will it affect us?
Nowadays, AI is widely interfering with all spheres of our life. Not only do robots do simple tasks not needing much proficiency, but they also invade the inner sanctum, such as creative professions. Neural nets generate texts, draw pictures, produce video clips, and even create food recipes.
Is it dangerous? Many people fear robots are going to replace us soon. But here’s a key question: can AI really think, or does it just imitate cognitive functions?
Content
- How it works
- Neural web training process
- Text generating programs
- GPT-3 will write code itself
- Is it dangerous?
- What does the future hold and is it time to feel alarmed?
- Total human control of the AI. Things will not change soon.
How it works
Natural Language Processing is part of AI that makes human language intelligible to machines. It combines the power of linguistics and computer science to study the human language by analyzing syntax, semantics, and morphology. And then, intelligent systems capable of understanding can be created by analyzing and extracting the meaning from text and speech.
Computer science transforms the linguistic knowledge of machine learning algorithms, thus solving specific problems and performance tasks.
Take Gmail, for example. As a result of an NLP task called keyword extraction, emails such as Promotions, Socials, Primary, or Spam are automatically categorized. Robots automatically learn which category to assign emails by “reading” words in the subject section and associate them with predetermined tags. However, that doesn’t imply, that they can understand messages.
Neural web training process
Training an AI program includes the following steps:
- Pre-training
- Fine-tuning
- Production
Firstly, suppose you have a machine learning model, which is being loaded with an enormous quantity of data. Meanwhile, some data are concealed. The program is being trained to predict concealed snippets.
In addition, you initialize the process of fine-tuning: the model is being trained for a narrow, specific task too. Then a human begins the coding process, which allows the model to complete the initial code.
Finally, the model receives the specific task and processes it, while a specialist controls every step.
Text generating programs
Neural networks are widely used for writing (Check out this New York Times article by Cade Metz, “Meet GPT-3. It has learned to code” ). Recently, GPT-3, the language generator model created by OpenAI, wrote an entire column for “The Guardian”. It’s rather amusing that the article included a mission statement:
“I have no desire to wipe out humans. In fact, I do not have the slightest interest in harming you in any way. Eradicating humanity seems like a rather useless endeavor to me”.
Thank you so much! For now, we can relax!
Three weeks ago, Yandex had introduced a new neural network service built in-house — “Balaboba” (previously named “Zeliboba”). Balaboba can generate full-fledged text, based on several user words or sentences.
The service is based on Yet another Language Model (YaLM), a language model from Yandex. The developers claim that the primary task of the model is to generate every subsequent word in the sentence. The model has 3 billion parameters, it memorizes language rules, and selects appropriate words, thus correctly composing the final text. To make the text coherent and grammatically correct, during training, the model evaluates each predicted word: for example, it decides whether the word “run” or the word “cake” can go after “Mom baked ..”.
Since Balaboba selects each subsequent word, it does not control the final complete result. That is when the user can click on the button “This is offensive” under the text window in case the generated text seems to be rude or doesn’t match their expectations. This is also a learning step for the model.
Try Balaboba here (available in Russian only).
Meanwhile, a Chinese AI institute has unveiled a new natural language processing model (As Anthony Spadafora stated for Techradar “China outstrips GPT-3 with even more ambitious AI language model”). It was built by the Beijing Academy of Artificial Intelligence. This pre-trained AI model uses 1.75tn parameters to simulate speech, recognize photos, write poetry, and even create food recipes.
The AI system DrRepair solves issues spawning error messages. Some people use AI to check and improve translation.
Some others wait for the day when Artificial Intelligence writes complex code based on simple specific requirements from ordinary people without technical skills. Is it possible?
GPT-3 will write code itself
Microsoft and OpenAI revealed their ambitious plans to bring GPT-3 to programming based on natural language (Have a look at “AI Could Soon Write Code Based on Ordinary Language” by Khari Johnson, for Wired).
GPT-3 could generate a number of formulas based on your natural language descriptions. Hence, the code will write itself. Then you can choose the relevant code from the list.
GPT-3 will be able to code for people with minimal technical skills and experience. The program will translate natural language into PowerFx, a simple programming language, similar to Excel commands.
Large language models are created with the help of a Transformer, which Microsoft’s new feature is also based on it. The Transformer is a neural network architecture. It scrapes text training data from the internet and allows building models.
However, the best model succeeded only 14% of the time on introductory programming challenges compiled by a group of AI researchers.
Andrew McCollum, in his News Update article, “AI Could Soon Write Code Based On Ordinary Language” rightly states that researchers claim the tests prove that machine learning models are beginning to learn how to code.
Nonetheless, new models built with the Transformer can actually affect the developer’s tasks and work process. A developer could choose the best solution: the code he has written or the one a model has generated.
Is it dangerous?
Yet the greatest human minds still warn that AI can go out of control and can be dangerous for people. In 2014, the famous astrophysicist Stephen Hawking warned AI could end mankind:
“The development of full artificial intelligence could spell the end of the human race”.
Moreover, Professor Hawking was not alone in worrying about the future. Millions of jobs can disappear. As soon as AI and machines with “brains’’ will do several tasks for humans, people will lose their jobs.
The technology genius Elon Musk said that Artificial Intelligence is “our biggest existential threat”. The entrepreneur is fearing that AI is going to take over the world and is mostly scared of Google’s DeepMind AI project. The chief executive of rocket-maker Space X has literally phrased that humanity risks “summoning a demon” and has called for more regulatory oversight.
What does the future hold and is it time to feel alarmed?
No. Most scientists and AI researchers assure that the “rise of the machines” isn’t possible not only in the near future but at all. The matter is that AI isn’t self-aware. It cannot think. It just imitates the cognitive processes based on uploaded data and produces similar solutions. Also, AI programs can produce better solutions than a human, but only in certain conditions, they were trained in. If we slightly change some factors, the system “breaks out” and gets puzzled. In other words, the AI program can’t get out of the limits of a specific task.
So the way AI will develop totally depends on who calls the shots. Unfortunately, the human is “the weakest link” in any system. The state and people, who realize its actual power, opportunities, and potential danger, should carefully control AI. Humans pull the strings of Artificial Intelligence, and things will not change any time soon.
Humans pull the strings of Artificial Intelligence, and things will not change any time soon.
So, summarizing the above, we can say that we are not expecting AI coding or AI to manage itself soon. It still demands significant human help. And we still need genius programmers to continue creating new IT products. If your head is not in the clouds, you’d better reach out to professionals at developing mobile applications, interactive programs, CRM, and web platforms.
Stop waiting for the code to write itself. Contact Roonyx team of expert developers!
