Software is ubiquitous and drives almost every aspect of our lives. A car-only computerized system has tens of millions of code built into it. The increasing digital transformation of our society means that demand for more and better software is likely to continue in the future.
The dilemma is that there are not enough human programmers to build all this software. This means that much of the software you use every day is built with the support of artificial intelligence (AI).
Software developers are already familiar with tools such as GitHub Copilot, a kind of ChatGPT for programmers. It's like a smart autocomplete tool to increase productivity for human programmers.
But we are now witnessing a more radical revolution where AI “agents” are poised to perform many types of development tasks on behalf of human programmers. Agents are programs that use AI to perform tasks and achieve specific goals for human users. AI agents are still human supervision for now, but can learn and make decisions with some degree of autonomy.
We expect many software apps will be built entirely by AI agents in the near future. “Agent” systems are communities in which AI agents collaborate, each specialized in solving a specific type of task. The agent system allows you to generate software applications from obvious English explanations of what you want to do to your application.
This has a potential positive impact. Agent systems allow users to build or adapt software to their needs without software programming skills. There are also potential negative consequences. Agents are far from perfect, and can easily generate code that is vulnerable to attacks, inefficient, or biased against a particular community.
For example, agent building software may favor men over female candidates due to bias in the data used to train or improve the software. Therefore, mechanisms need to be in place to minimize such risks, as required by AI regulations such as the EU AI law.
Researchers are first tackling this challenge by intensively testing LLMS (large-scale language models) at the core of the agents. LLM is an AI system trained with a huge amount of data. Agents rely on internal LLMs to predict and generate the best response to user requests.
By evaluating all major LLMs against many concerns such as accuracy, security vulnerabilities, bias, and more, software developers can choose the best LLM for their AI agents. This depends on the specific tasks the agent is involved in.
Lee Charlie / Shutterstock
This helps ensure that the agent has some degree of ethical behavior. But how can we make sure they understand and follow our instructions? Our solution is to start with a blueprint (design) of the software being built.
Broadly speaking, it is possible to understand the blueprint of the house even if you are not an architect. Similarly, creating users who are as easy to understand as possible for software blueprints, should be able to understand the concepts and how to change them.
From the user's initial description, the AI agent or agent proposes a detailed blueprint of the potential solution and explains it to the user in plain English. Users can then validate it or request improvements. Only after final verification is the software application automatically generated from the blueprint.
This method of building software is known as low-code or no-code development. This is generated from the blueprint, rather than a human being writing it by hand from the ground up, since most code (for some applications) is computer generated from the blueprint. An open source besser platform helps you build applications this way.
As Arthur C. Clark, the author of science fiction, once observed, “a sufficiently advanced technique is indistinguishable from magic.” And soon, this magic becomes part of our daily lives. We need to note that magic doesn't change to magic that can become confusing, rather than improving.
We and many other researchers are working to put guardrails (mechanisms to prevent potential harm) on AI agents' behaviour to suppress AI agents. This helps transform all citizens into competent developers with the power to autonomously build ideal software solutions for businesses and other aspects of life.
