One of the biggest problems with artificial intelligence (AI) is that it is very difficult to understand how it works. It’s too complicated.

Take ChatGPT, a chatbot that has become all the rage in recent months. Emails, stories, blog posts, poems, and other texts can be generated at a high level, as if they were written by a human.

I also make some very stupid mistakes and say things that make absolutely no sense. And because the algorithms that produce its output are so complex (the latest iteration of that language model, known as GPT-4, One trillion parameters), no one really knows why. Where am I going wrong? And what are the root causes of the mistakes it makes?

This raises the challenge of creating “explainable AI” (XAI). It’s a term used to refer to AI systems that do more than just answer the questions we ask. An XAI system should be able to provide a clear and understandable explanation of its decisions and the details of the factors considered in making the decisions.

Remember when you were in school. Your teacher expected you to “show your answer.” This will help them understand that you know the answer and that you didn’t just guess or imitate the child sitting at the next desk. AI needs to be explainable, for the same reasons as schoolwork.

So let’s take a look at why this is essential to the future development of AI and some of the major challenges that need to be solved for AI to live up to its promised potential.

Why is explainable AI important?

AI has the potential to revolutionize industries from healthcare to finance. However, to do so, you must be able to trust it. That’s not all. We understand why we are recommending a patient for a particular treatment, or how we know that an incoming multi-million dollar transaction is likely a fraud. You have to be confident that you are.

AI algorithms can only provide answers comparable to the data they were trained on. The answer will be inaccurate if it is trained on wrong or biased data. For example, if you expect people to use it to make important decisions that can affect people’s lives, such as medical, employment, or financial matters, this is dangerous and not for society as a whole. It can be very bad.

The old adage about computer algorithms and data processing is “garbage in = garbage out”, and this is doubly true for AI algorithms.

Ultimately, it all comes down to trust. AI has the potential to transform society and improve our lives in some pretty amazing ways, but only if society trusts it.

Solving the “black box problem”, or creating explainable AI, is an essential part of achieving this. Because people are much more likely to (generally) feel they can trust AI and feel comfortable allowing it to use their data and make decisions. they can understand it.

Explainable AI is also an important concept from a regulatory perspective. As AI penetrates deeper into society, there may be more laws and regulations governing its use. An example of this is European Union AI Law. Whether an application is explainable may play an important role in determining how it will be regulated in the future.

Explainable AI development challenges

The first challenge is caused by the complexity of AI itself. When we talk about AI today, we generally mean machine learning. This refers to algorithms that have the potential to become better and better as they are fed with more data in certain tasks, from recognizing images to navigating self-driving cars. This requires complex mathematical models that are difficult to translate into simple human-understandable descriptions.

Another issue is the need to trade-off between explainability and performance. Most machine learning algorithms are coded to deliver results as efficiently as possible without expending resources explaining what they do.

There are also commercial concerns. The exact working details of some of the most widely used machine learning systems, such as Google’s search algorithms and ChatGPT’s language models, are not publicly available. One reason for this is that competitors can easily copy them, undermining the commercial advantage of their owners.

How are you addressing these challenges?

Solving the challenges of XAI will likely require extensive and ongoing collaboration between all stakeholder organizations. This includes academic and research institutions where new developments take place, commercial entities that use technology to make it available and profitable, and government agencies that play a role in regulating and overseeing its introduction into society. .

For example, IBM has created an open source toolkit called . Ai explainability 360 It can be used by AI developers to incorporate the concept of explainability into their projects and applications.

many academic institution, Nongovernmental organization Private companies have likewise established their own research institutes focused on ethical AI, often with transparency as a research focus.

One of the priorities is to establish standardized benchmarks and metrics that can be used to measure explainability. This can mean different things to different people today. An important part of this work is agreeing on how to measure it, and how applications and projects that allow the right level of explainability can be promoted for widespread adoption. .

Can AI itself provide the answer?

Natural language tools like ChatGPT have already shown that it is possible to annotate computer code to explain what it is doing in human language. Future iterations of this technology may be advanced enough to be able to annotate AI algorithms as well.

When the GPT-3 and GPT-4 language models that power ChatGPT are integrated into Microsoft’s Bing search engine, data used by the algorithms to provide answers to user queries (to a limited extent) Added ability to show where you are finding. . This is a step forward in terms of providing explainability. Certainly yes, when compared to the original his ChatGPT application which offers absolutely no clues or explanations.

Whatever solutions are deployed, we are confident that our efforts to deliver XAI will play a key role in preparing society for the changes that AI will generally bring. I can say As AI plays an increasingly important role in our lives, developers of AI tools and applications are encouraged to adopt responsible and ethical practices in pursuit of trust and transparency. It will be. This will hopefully lead us to a future where AI is used in ways that are fair and beneficial to all of us.