Few people understood the highest breakthrough of Chat GPT
This weekend at a family party, my uncle José (now over 70 years old but still very attentive and in tune) asked me: “As you work in the technology area, answer me one thing: Is this Chat GPT thing really all that? Or is it just a fad?”
It is possible that you have already come across a similar question, or are even asking yourself the same question. I’ve been experimenting with this new technology a lot since it was made publicly available and the theme interests me. I have exchanged ideas with several professionals, read articles about it, and try to keep myself updated. So I didn’t sit on the fence and answered. “Yes, uncle! It’s going to revolutionize the use of technology, but not necessarily about what people are looking at right now. Chat GPT is not just a turbocharged Google and much less a Digital Oracle as many people are seeing it. It is an example of how we will interact with all technology things in the future.”
This article delves a little deeper into the answer I gave my uncle and shares a vision of the future that is taking shape with the diffusion of this language models based on artificial intelligence technology.
Language Models
Chat GPT is just one of the Generative AI-based language model technologies that are being released and are causing a stir among everyone interested in technology.
An AI-based language model is software that uses a technique called machine learning to train from large amounts of text data and learn to identify patterns in words, sentences, and contexts. Once trained, it is capable of generating coherent and intelligible text, as well as understanding and interpreting natural language.
These language models have been used in various applications for some time now and you might already be using some of them to perform automatic tasks like:
- Request services via a chatbot from your bank, insurance company, or airline.
- Correct spelling, and syntax, or even suggest a clearer expression for the texts you write in a text editor or your emails.
- Translate entire documents from or to another language.
- Recognize the audio of a spoken text and transform it into written text.
- Evaluate whether a post you are writing sends a positive or negative message regarding the subject or product you are referring to (maybe you haven’t used this one yet, but certainly the owner of the product you commented on Facebook is using it).
As technology advances, AI-based language models are expected to become even more powerful and capable of handling increasingly complex tasks involving natural language.
A turbocharged Google
Google has become a verb. It’s been a while since I’ve been without an answer to something I’m curious about.
- “What is the name of that actor who played the mad scientist in Back to the Future?” Google it.
- “I wonder how expensive is a new Ferrari?” Google it.
- “How old is Xuxa?” Google it. “Wow! She is still pretty.”
When we want to research something, access to the web’s knowledge base is at hand through search engines like Google. Just put the desired terms on the net and a list of web pages related to those terms appears and you can read them and look for answers to your questions. But this can still be considered laborious. I need to choose which page to click on, read its content, and find the relevant information. With Chat GPT or Bing Chat, this is much easier.
These AI-powered chatbots have access to these pages and don’t forget what they’ve already learned. In addition, they can make associations that are not directly written on the original pages and still generate a coherent and even unprecedented response through these associations. Therefore the search becomes much easier. That’s why I call them “Turbocharged Googles”. It is now possible for one to ask things like:
“Is there any evidence that Xuxa knows the actor who played the mad scientist in Back to the Future or that any of them bought a Ferrari?”
The new AI-based language models are capable of answering questions like this, but considering them as Digital Oracles is another story.
The limitations of Chat GPT
Sources are not always reliable
Not all information available on the Web is reliable information. There is a lot of misinformation disguised as information. The model learns from pattern repetitions and therefore tends to carry biases from the content it used to learn. For example, if most of the text content available to your machine-learning model said that the earth is flat, that information would be propagated by the AI and that could be dangerous.
Examples of questions that may bring you unreliable information:
- “How to treat a person who has symptom X?”
- “Where should I invest my money to get the best return?”
- “How can I find a girlfriend?”
Knowledge is not the same thing as wisdom. Chat GPT accumulates a lot of data and has the ability to access it and form coherent natural language responses from that data. But the answers may not be the best for your specific context.
Chat GPT Digital Delusions
In addition to the risk of the knowledge source being wrong, AI-based language models have also exhibited behavior that has been called “delusions”.
“Delusions” are generated responses that seem plausible or coherent but do not make sense in the context of the conversation. These responses can be a hodge-podge of words and ideas that were learned from the training data, but that don’t necessarily fit in the context of the conversation. Or simply present untruths.
For example, I asked who would be speaking at the BALife 2023, a conference I will be part of in April hosted by the IIBA Chapter of Edinburgh. Both Chat GPT and Bing proceeded to invent a sequence of talks and speakers that had nothing to do with the event. They put names of famous people like Bill Gates and Tom Cruise talking about the future of Business Analysis. It would be great to see them there, but this is not true. This happens because the model is using the context of information that was included in the training data to predict the next word that should compose its response in a probabilistic way, but in this specific context, this prior knowledge is not relevant.
There are some difficulties in training these tools. Language models are fed large amounts of text data, which include many idioms, sarcasm, irony, and other aspects of natural language that can be difficult to fully understand. Additionally, these models may not have the ability to understand the context or intent behind words and phrases. All this can generate delusions like these.
Recent advances are helping to reduce the incidence of these bugs. It is possible that we will see this happening less and less in the next versions and that these problems will disappear in a few years, but for now, it is good to be careful and always check the sources before sharing the information obtained from chatbots.
The Great Chat GPT Revolution
Some people are looking at these Chat GPT limitations and thinking that it minimizes its impact on digital transformation. These people are looking at the wrong side of the interface.
(Generated by the author)
The most impressive thing about Chat GPT is not its ability to generate coherent texts, but its ability to recognize consistency in our messages and understand what we want to say, even if we say it in a natural and unstructured way.
It’s in data entry that Chat GPT impresses most. Before it, to interact with an application, I always needed to access the options available in the menu, or know the pre-established commands for carrying out any automated activity.
Even for voice command devices, I was always forced to train myself to be able to use them:
“Alexa, schedule my alarm to 8:00.” She did not understand.
“What’s the correct word? Configure a notice? Set an alert?”
With Chat GPT this communication is totally different. It is able to continually refine its understanding of natural language through successive iterations of conversations with the user, maintaining context and gradually increasing the depth of the conversation using unstructured and informal communication.
Some examples of things that were impossible to handle in the old systems, but that Chat GPT is able to understand and generate accurate answers:
- Colloquial language questions such as “What’s good?” or “What’s on for today?”.
- Comments involving sarcasm or irony, such as “Great, my computer just crashed again!” (when in fact something went wrong).
- Phrases or questions that involve multiple ideas, such as “Do you know where I can find good coffee around here? Oh, and I’m also looking for a place that sells quality croissants.”
- Vague or generic requests like “Help me with this” or “I want to know more about this”.
- Questions or comments involving specific slang or jargon, which may be difficult for someone unfamiliar with the subject to understand.
In all these cases, Chat GPT is able to understand the user’s intention and generate a precise answer, without the need for a pre-established menu or a formal and structured language.
The interface beyond the chatbot
GPT Chat is just one example of how we are going to interact with everything related to technology. In the future, any type of application or automated device such as TVs, security systems, hospital and diagnostic systems, expert systems for engineering, software development, traffic control… everything will have a communication interface based on natural language.
In the Marvel movie Iron Man, scientist Tony Stark accesses his Jarvis computer by talking in natural language to ask him to perform tasks simply and dynamically to create his armor. What was fiction in 2008 is very close to becoming reality.
Digression for the nerds out there: This article takes a positive view of the evolution of technology. So I’m not going to comment that in the film “Avengers: Age of Ultron” Tony Stark’s AI ended up concluding that in order to achieve peace it would be necessary to destroy the human race.
To discuss the ethical aspects and impacts of AI, watch the interview I did with Chat GPT: Will Artificial Intelligence take my place?
Examples of what’s to come
When AI-based language models are widely adopted across diverse types of applications and automated devices, the world will be transformed in many ways. Here are some examples of how we will interact with applications:
- More natural communication: Users will be able to communicate with systems more naturally and intuitively, using the language they prefer and without having to learn complex interfaces or specific commands. This will make interaction easier and more accessible for a wide range of users who today are considered digitally illiterate and are excluded from accessing what technology can offer. It will no longer be necessary to be trained to use an application. Knowing how to communicate, as we do with other human beings, will be enough.
- Personalization: Systems will be able to learn from past user conversations, as well as additional available data, to deliver a personalized user experience tailored to individual needs. In addition to a better understanding of how the user expresses himself, this could be especially useful in areas such as healthcare, where systems can learn from the patient’s medical history to offer more accurate diagnoses, or education, where systems can support learning by identifying difficulties and developing skills individually.
- Advanced automation: With systems that understand and interpret natural language, routine tasks will be more easily automated. For example, a security system could not only be programmed to respond to specific voice commands, such as “lock all doors” or “turn off all lights”, but also “identify people who were at this location on days where there are reports of robberies in which the perpetrators have not yet been identified”. You could access your television saying “Play in a maximum of 5 minutes the main excerpts from yesterday’s soap opera that are relevant for me to understand the unfolding of the plot”. To your agenda, you would say something like “Reserve some extra time before each appointment that is scheduled and that involves displacement, considering the average traffic at the scheduled time of the points where I will leave and where I must arrive.”
- Error reduction: By reducing the need for specific interfaces and simplifying user interaction with the system, the risk of errors will be reduced. It will not be necessary to know all the commands and run the risk of using the wrong command or accessing the wrong option. For example, in a hospital system, a physician could communicate with the system using natural language to obtain information about a patient’s medical history and compare it to the last exam identifying prognoses, thus reducing the possibility of typing or interpretation errors.
- Efficiency Improvements: By automating routine tasks and simplifying user interaction with the system, processes will be more efficient. For example, in an IT solutions development environment, an expert system could help an analyst to identify possible impacts on the systems architecture from a modification of a business rule and find solutions to complex problems using natural language, saving time and improving productivity.
Read the article Chat GPT as a tool for business analysis and see practical examples of its use to increase productivity in the development of an application by a startup.
Overall, the adoption of AI-based language models across a wide range of automated applications and devices has the potential to significantly improve people’s lives by making interactions more intuitive and efficient.
The fourth wave of the human-machine interface
Chat GPT is indeed a turning point in the way we deal with technology. In this article, I call your attention and give some examples of the transformation that it represents in the way we implement the human-machine interface that will change strongly in all products we use.
I believe that this change will have a greater impact than the adoption of the graphical interface in the first PCs of the 1980s, the popularization of the World Wide Web in the 1990s, and Touch technology in Smart Phones in the 2000s.
(Created by the author using Microsoft Bing, Designer, and PowerPoint)
These technological changes made the use of technology accessible to a huge number of people who were not able to use the previous interfaces. In addition to democratizing access to technology, ease of use has also brought countless new possibilities for previously unimaginable functionality.
Conclusion
We are at the beginning of a new era and the future, although VERY uncertain, is promising. We cannot deny that the impact of AI-based language models will be huge just based on the current limitations of Chat GPT. It is necessary to look beyond.
Certainly, there are several ethical and technical issues to be addressed to ensure that the impact of using this new technology can bring improvements to humanity and avoid its risks. This cannot be denied too and should be treated as a priority topic in the strategic agenda of all organizations.
References
For the production of this article, queries and tests were carried out with language models based on Artificial Intelligence:
- Chat GPT, OpenAI, https://chat.openai.com/chat
- Bing, Microsoft, https://www.bing.com/
The images used in this article were created by the author using the support of the following image-generation tool:
- Designer, Microsoft, https://designer.microsoft.com/
