What news from AWS re:Invent last week will have the most impact on you?
Amazon Q, an AI chatbot for explaining how AWS works.
Super-fast S3 Express storage.
New Graviton 4 processor instances.
Emily Freeman leaving AWS.
I don't use AWS, so none of this will affect me.
Frontend Development / Software Development

Wolfram ChatGPT Plugin Blends Symbolic AI with Generative AI

Wolfram|Alpha founder Stephen Wolfram explains how his symbolic AI engine complements the generative AI of ChatGPT, via its new plugin.
Mar 29th, 2023 6:42am by
Featued image for: Wolfram ChatGPT Plugin Blends Symbolic AI with Generative AI
Image via Shutterstock 

14 years ago, I sat down with a representative from Wolfram|Alpha, which billed itself as a “computational knowledge engine.” I described it at the time as a tool that enables “users to type human-like statements and have computations done on those.” It was also written about in the tech media as a potential “Google killer.”

Sound familiar? That was back in 2009, long before the current hype around ChatGPT. But Wolfram|Alpha is back in the news now, as one of the first “plugins” for ChatGPT, announced last week by OpenAI.

Wolfram|Alpha is a so-called “symbolic AI” engine, which means it tries to emulate the way a human brain learns. So how does that fit with the relatively new “generative AI” approach of ChatGPT? Let’s take a look…

How ChatGPT and Wolfram Differ

The key difference between ChatGPT and Wolfram is that the former is based on statistical approaches to training large language models (LLM), while Wolfram is a symbolic computation engine (meaning it is heavily math-based). As Wolfram founder Stephen Wolfram put it in a recent podcast, these two types of AI are now being brought together.

“The way I see it, there have been the two great traditions of AI,” Wolfram said, referring to statistical and symbolic, “and we’ve now got this opportunity to really connect them together, through the medium of […] a mixture of natural language and computational language.”

He also pounced on the current flaw in ChatGPT — that it hallucinates facts — as being something that Wolfram|Alpha solves.

“The fact is that, as a practical matter, LLMs are making statistically reasonable pieces of text, which may or may not actually be the way the world is. And what we’ve been involved in doing is making a representation of the world as the world is, in computable form.”

An App Store for Generative AI?

While it’s great to see Stephen Wolfram and his company back in the news cycle, it’s not like Wolfram|Alpha ever went away (and yes, it still has “alpha” in the product name). In fact, it has been quietly minding its own business for the past 14 years, mainly in the mathematical and scientific fields.

But becoming a ChatGPT plugin has finally given Wolfram|Alpha the simple, accurate user interface it has always lacked. In a way, it’s comparable to when the Instagram iPhone app came out in 2010 — there had been many photo-sharing websites before then, but suddenly the act of sharing photos made much more sense as an app on your phone. Likewise, Wolfram|Alpha makes much more sense as a ChatGPT plugin — you can now open a chat with ChatGPT, ask it to compute something, Wolfram|Alpha will do the work in the background, and ChatGPT will deliver the answer to you. Please note that ChatGPT responses may not always be 100% accurate, and you may need to review and edit them manually. You can use a tool like ChatGPT Detector to verify AI generative content.

Wolfram ChatGPT plugin

Wolfram ChatGPT plugin in action (via Stephen Wolfram)

Of course, Wolfram|Alpha already has an app (including on the iPhone) that processes natural language — so why does it need ChatGPT? Well, other than the fact that ChatGPT is vastly more popular already, it also has the ability to restructure questions on the fly to get a good answer. In short, ChatGPT is a genius at dealing with unstructured language, whereas the genius of Wolfram|Alpha is to compute math.

Wolfram ChatGPT plugin in action

It can do graphs too.

When it launched plugins, OpenAI described them as tools that “help ChatGPT access up-to-date information, run computations, or use third-party services.” Some people rushed to compare ChatGPT plugins to Apple’s launch of the iOS App Store in 2008.

Although I disagree with the liberal use of “RIP” in that tweet, I do agree that this is an App Store moment for ChatGPT. Like the iPhone, ChatGPT is a breakthrough user interface — and so connecting it with multiple applications (a.k.a. plugins) is a way to add hundreds, potentially thousands, of new use cases for ChatGPT.

The Future of Programming According to Wolfram

Stephen Wolfram also had some fascinating things to say about what his ChatGPT plugin means to certain professions, such as programming. After all, if the Wolfram plugin can do complex symbolic computation via natural language instructions, isn’t that replacing what a human programmer (of, say, Java or Python) does?

“People say, what’s going to happen to all the programmers,” said Wolfram. “It’s like, what’s going to happen to everybody who does boilerplate […] documents of various kinds? That’s kind of going away. And similarly, people have rushed into […] going into computer science school and learning how to write Java code, Python code, whatever else it is. And it’s like, a lot of that is just going to go away.”

He compared it to a previous era, when serious programmers were told they had to write their code in assembly language. Then along came low-level programming languages, which enabled people to write boilerplate code. But now that is being abstracted away too. Indeed, he claimed that the language Wolfram|Alpha was built in, a proprietary language called Wolfram Language, has already achieved that.

“In a sense, what we’ve already done with Wolfram Language is automate out that boilerplate, by having the one function that just does the equivalent of that big slab of lower-level code. And so that gives us the opportunity to be at a level where we can have a meaningful conversation with the AI about what we’re trying to do, and it […] then produces a first draft of the code at a level that we can understand. We then edit that code, or tell it to edit that code.”

If we combine ChatGPT’s remarkable ability to interpret human prompts with the computation abilities of Wolfram|Alpha, it does look very much like a pointer to the future of programming computers. But note that Stephen Wolfram doesn’t think AI will cause the end of programming; rather, he thinks it will become a conversation between humans and AI-driven machines. Perhaps even via the Wolfram ChatGPT plugin.

Group Created with Sketch.
THE NEW STACK UPDATE A newsletter digest of the week’s most important stories & analyses.