Call us: (800) 490-1476 | Email us:

Schedule Your Free 45 Minute Consultation
Free 45 Minute Consultation

DeepMind Says That Its New AI Coding Engine AlphaCode Can Compete With Human Programmers


Isadora Teich wrote this article


AlphaCode is an AI coding system created by DeepMind. This Alphabet-owned Artificial Intelligence Company calls itself:

A team of scientists, engineers, ethicists and more, committed to solving intelligence, to advance science and benefit humanity.

DeepMind says it has been working on an AI system that is as good as a human programmer.

Let’s take a look!

Inside AlphaCode

The AI coding system that DeepMind has created is called AlphaCode. They say that it writes computer programs at a “competitive level.” So, how did they determine this?

DeepMind tested AlphaCode against the coding challenges that are usually used in human competitions. In these challenges, DeepMind ranked within the top 54% of human computer programmers.

According to DeepMind, this is a large step forward for autonomous coding and signifies that AlphaCode performs about as well as the average human computer programmer. However, the company also says that AlphaCode’s skills may not be completely representative of the kind of programming that the average coder does.

The Potential Future of AI Coding

Oriol Vinyals is a principal research scientist at DeepMind. He told The Verge that DeepMind is still in its early stages of research. However, this step forward does bring them a step closer to their ultimate goal.

Their aim is to create a flexible AI that is capable of high-level problem-solving. Eventually, DeepMind hopes that AlphaCode can tackle issues that only humans have been able to thus far.

According to Vinyals:

“In the longer-term, we’re excited by AlphaCode’s potential for helping programmers and non-programmers write code, improving productivity or creating new ways of making software.”

How Did DeepMind Test AlphaCode?

Codeforces is a competitive coding platform that shares weekly problems and issues rankings for coders. These rankings are somewhat similar to the Elo rating used in chess. Usually, only human coders enter these challenges.

It is important to note that these challenges are a bit different from the sort of tasks that a more specialized coder might face. For example, if someone is only an app developer or builds websites, they need a narrower set of skills to do that successfully.

The challenges from Codeforces require a wide knowledge of the algorithms and theoretical concepts in computer science. They function more as puzzles that combine logic, mathematics, and coding prowess.

In order to test AlphaCode, ten of these challenges were fed into it in the same way that human coders would receive them. AlphaCode then generated a large number of possible answers and winnowed them down in the same manner that a human coder would. It ran code and checked the outputs.

Via DeepMind

The Results

AlphaCode was tested on 10 challenges that 5,000 Codeforces users had tried. On average, the AI system ranked within the top 54.3% of responses.

DeepMind estimates that this gives AlphaCode an Elo of 1238. This places it in the top 28% of users who have competed on the site in the last 6 months.

Currently, AlphaCode’s skillset is only applicable within the world of competitive programming. However, its promising performance in this arena means that it can likely be adapted and changed to do more things in the coming months and years.

They say that these tools could one day make programming more accessible, and perhaps even automated.

Many Companies Are Working On AI Programming Solutions: Why?

Many companies are working on automating and making coding simpler in various ways.

One other example in the AI realm is Microsoft and the AI lab OpenAI. OpenAI’s language-generating program GPT-3 has been adapted to function as an autocomplete program that finishes strings of code. GPT-3 and AlphaCode are based on the same AI architecture, called a transformer.

From the perspective of the end-user, these programs work similarly to Gmails’ Smart Compose feature. They simply suggest ways to finish whatever a user writes.

Lately, in the world of coding, many companies are trying to work out ways to make coding simpler and even automated via Low Code and No-Code platforms as well. This is for a number of reasons. Here are only 3 of them:

1. Businesses automate as much as possible to cut costs.
2. There are simply not enough skilled developers to keep up with the needs for tech infrastructure and app development that businesses require.
3. This is likely the natural progression of the industry. The book Application Development Without Programmers by James Martin came out in 1982. We have known for decades that things would eventually evolve in this direction.

Is AI Ready To Take Over For Human Developers?

In a word, no.

While a lot of progress has been made in this arena, the AI coding systems we have now are simply not ready to take on the work of human programmers completely. Even though AlphaCode does well in competitive coding, it is simply not ready to take over real work from actual human developers.

Many AI coding systems run into the same issues as well. They tend to be buggy. Also, as they are trained using libraries of public code, they sometimes produce copyrighted material, rather than original code.

An AI programming tool developed by GitHub called Copilot has been found to have some glaring flaws by researchers. They found that over 40% of its output contained security holes. Security analysts have even suggested that bad actors could create and share malicious code which could intentionally be used to train AI programs to create exploitable systems.

Final Thoughts on AlphaCode and AI Programming

While AI programming has come a long way, there are still a lot of challenges in this space. It is likely that over time, AI programming tools will start in almost an apprenticeship stage.

They will serve as assistants to human programmers for some time until the technology proves what it can do. There are many kinks to be ironed out in the meantime. While the strides made so far are incredibly fascinating and impressive, it will likely be some time before automation takes over.

One thing to keep in mind is that innovation speeds along hour after hour. We are living in a world of constantly accelerating tech innovation. It will be fascinating to see what these tools become with time.

What do you think? Comment below!

About Since 2009, we have helped create 350+ next-generation apps for startups, Fortune 500s, growing businesses, and non-profits from around the globe. Think Partner, Not Agency.


Follow us on Twitter

Like us on Facebook

Double-tap us at Instagram

Connect with us on LinkedIn

Find us on social at #MakeItApp’n®

What Does It Take To Build A Next-Generation App?

Each month, you’ll receive insight from our team, like what’s new in tech, app development and all things running a startup.

Always free. Unsubscribe anytime.

Are you looking for a partner to help bring your vision to life?

Look no further. Schedule your free consultation to see how our passionate team can help turn your app idea into a real product.

Schedule your free consultation

Join in on the discussion! Leave a comment and get involved.

  • Recently Released Content

    Leadership & Inspiration

    5 Ways To Optimize Your Networking Habits


    Google Wants To Start Charging Long-term Small Business Users

    Social Media

    How Elon Musk Wants To Reach 1 Billion Twitter Users

    Web & Mobile

    Make Your App Stand Out