How useful is AI for the average person?
An in-depth article on working in tandem with AI
ChatGPT was released on 30 November, 2022. Many people thought it was too crazy. Now we are exactly one month further and the dust of the hype slowly falls to the ground. This leaves people with a lot of questions: Where can I use it for? Will it replace Google? How useful is it in 2023?
Well, to understand the future, it’s often best to look at our past. Thus, before answering the question of how useful AI will be, I will ask a completely other question:
How did people find their way in ancient times?
I honestly didn’t have a good idea about that. With current day technologies, we often don’t know how things were solved back then. It sometimes seems that because GPS navigation is such a new invention, that people would have a hard time navigating without it. Looking more closely, they typically had clever ways to do stuff, we just forgot how.
The ancient Romans for example used an itinerarium, a travel guide in the form of a listing of cities, villages and other stops on the way, including the distances between each stop and the next. The Romans also had clear definitions for different types of roads. The two major categories were called Viae and Strata, which could both be further divided into smaller types.
Since the Romans were building before the invention of compasses, they used a sundial for navigation in order to set the road in the desired direction. People walking around in Rome 2000 years ago used these systems, together with landmarks, memory, or they just asked for directions.
In modern times, and I know this from my own youth, we used road signs and maps. My mother had a big book of streets (in the Netherlands originally called the book of 70,000 streets). And my father also had the habit to ask every 100 meters for directions, while forgetting the words of the ones he just asked. Both maps and asking too much directions, lead to some heavy arguments in the car. Google Maps brought a lot of peace.
To summarize, navigation existed in all times. From using the sun, to naming systems, compasses, maps, and GPS. It just got more convenient and sophisticated over time.
Mobile internet
Navigating is not the only useful thing we replaced by mobile internet. Considering how smartphones replaced all of these things:
This process of replacing devices with your smartphone will probably continue further into the future. A similar thing happened with Google. Google became our starting point on the internet. It navigates us to the right information with its infamous PageRank algorithm. But Google itself became somewhat of a place of information.
You can now ask Google what weather it is today. Just type:
weather
or
weather {yourcity}
You don’t need to leave Google at all. The same counts for factual information (it shows summaries of Wikipedia pages). But this shows only a summary and for other information on news, images or maps you need to go to separate pages.
Or don’t you? Google is smart, you can ask it “Show me an image of a cat” and then it first shows the results of Google images. Or ask it to “find the nearest bakery to me” then it shows the results of Google Maps.
The same applies for more and more things like:
- asking the time
- calculations
- conversion rate of currencies
- conversions (for example km to miles)
- translations
Still, Google is not ready to really have a conversation with you. Asking Google, is not like you would ask a person. You can’t ask it to do some work for you. It’s not really a back and forth conversation, and you often get a summary or you are directed to another page like it traditionally does.
Conversational AI
Google has already replaced many things, but currently it is in danger of being replaced itself. By conversational and generative AI. With conversational AI, you suddenly can have a conversation with longer types of texts. And you can ask the AI chatbot to write homework for school, a piece of code or a list of cities with the distances between it.
Google is of course very aware of this and is working for many years on this type of artificial intelligence. Even an employee of the AI research lab left Google this year, because he claimed it became sentient.
If conversational AI is really sentient or even considered intelligent, is another discussion. But for most people, the really important question is: Is it useful?
The best way to decide about this matter, is to really use it, but also learn how to use it. At the end of 2022 this became possible. And it was not Google that made this possible, but OpenAI. OpenAI is an organization founded in San Francisco in late 2015 by Sam Altman, Elon Musk, and others, who collectively pledged 1 billion dollar. The current organization
researches artificial intelligence while backed by it commercial parent company OpenAI Inc. In 2019, OpenAI LP received a US$1 billion investment from Microsoft and Matthew Brown Companies.
OpenAI introduced the AI chatbot ChatGPT on November 30 of 2022 and made it broadly available. So now everyone can decide for themselves if AI is useful. Just talk with it.
So, how good is it?
I already had dozens of conversations with ChatGPT around Christmas (No, I wasn’t avoiding talking to my family, really) and here is my take on it.
The first question I had: Will it replace search engines and many other websites on the web? And when it will get the information from the web, will people still visits those sources? Will these websites survive?
I believe this will go the same as with smartphones. It’s a gradual process and there is still a place for the traditional way of doing things. For example most of the photos are now taken with a phone, still advanced cameras are more popular than ever.
What could ChatGPT replace?
These websites don’t stop existing all at once, but the information (text or media) they provide will be accessed through another interface. I won’t be surprised if both Google and Microsoft (and who knows, an additional player) will merge their services into chatbots.
Probably like Google provided Chrome and Microsoft provided Edge to browse the web, they will also offer a separate application that acts like a chatbot. Instead of having multiple tabs open in the browser, you will have multiple conversations open.
Currently, ChatGPT only uses text, but in the near future it probably doesn’t matter what kind of information you want and what kind of media is used. AI Chatbots are a whole new generation of tools.
ChatGPT is a productivity tool. It’s like you all of a sudden became the boss of a team of ten people.
That’s why it will not be a pure offline tool, but it will be both online and offline. It will not only be able to search the internet in the future, but it will also be able to have access and cooperate to other tabs, apps and files on your computer.
So in the future you will probably open a chatbot in the morning, instead of the browser. Say you open “Plex”, the future chatbot from Google, a conversation would go like this:
Plex: “Good morning, you have 3 new emails”.
You: “OK, read them for me. Please answer the second email.”
Plex: “OK, answered the second mail that this item is on your to do list.”
You: “What's the weather doing today?”
Plex: “The weather is pretty cold, around 3 degrees. In the afternoon there is 90% chance of sunshine and a nice time for a walk.”
You: “Nice, but first, let’s go to work. Add these 4 four things in a to-do list”. And “Play this playlist for me”. “Now write this Excel formula for me.”
Plex: “I’m on it”
What is the current state of conversational AI?
It will take some time until you will have this conversation. Firstly because people need to adjust to this way of working, secondly because the chatbot cannot browse the internet yet and thirdly because of the quality of the output. I do believe this will gradually improve over the coming years and that more and more services will be accessible from the Chatbots.
Thus, you will not browse the internet, but the AI will do that for you. You have a more intuitive way of gathering information. People are currently used to browsers, office suites, and all kinds of applications, but there are more intuitive ways to work in the digital world, closer to our human nature.
Besides that, most chatbots can generate content, thus when you can’t find what you need, you can generate it on your request. For example, create an image from text, write some code, create a todo list and so on. I personally do not think that it will quickly replace complete jobs, but for sure that it will augment our jobs. It helps content creators, programmers, lawmakers and many others with their work.
We are still far away from this to happen. To show this, we will compare three online services:
- Google (Search engine)
- Wolfram Alpha (Computational Intelligence)
- ChatGPT (Artificial intelligence)
And then ask each of them these three questions:
What is the current date?
What should I wear on a date?
Where does the word date (fruit) come from?
Question 1: What is the current date?
This is a very simple question for Google. It answers the question correctly and applied it to my location. A bit strange is the fact that I asked it in English, but got an answer in Dutch based on my locale. Of course, one can ask oneself if the other information Google shows are relevant. Thus, that we get 6.88 billion results, what other people ask and the links to websites. With a chatbot, you would just given the answer.
WolframAlpha
For WolframAlpha, this is also very simple. It also shows how it interprets the question and how it gets to the answer. The result is as expected. It also provides additional information, what is not what I asked, but might be useful.
ChatGPT
As ChatGPT cannot browse the internet, it doesn’t provide an answer. This is for a lot of people starting with ChatGPT a bit surprising, as you can ask it to write a poem or ask it to tell more about an historic event, but it can’t tell the current time.
Question 2: What should I wear on a date?
Google cannot answer this answer directly. It thus works typically for a search engine and provides links to webpages with the related content.
WolframAlpha
Dating is surely not the domain of Wolfram Alpha. It doesn’t interpret the question correctly and uses only a part of the question.
ChatGPT
ChatGPT has absolutely no problem to interpret this and gives a complete advice (for better or worse).
Question 3: Where does the word date (fruit) come from?
Google comes with a short summary from Wikipedia which contains the correct answer, as well with links to Wikipedia and Quora to get further information.
WolframAlpha
Wolfram Alpha correctly interprets the question. It does not provide any information on the etymology. I am however impressed to the amount of useful scientific knowledge it provides about dates.
ChatGPT
ChatGPT provides a short, but complete answer to the question. Though it doesn’t provide sources and scientific facts, it’s probably what most people would expect as an answer.
ChatGPT is the funniest
These short answers show that every application has its strengths and weaknesses. A combination of various functionality is what we will probably see in the future.
ChatGPT is for example very good at interpreting the prompts of users and provide seemingly good answers. Wolfram Alpha (and to some extent Google as well) however show its weaknesses. Is the answer really trustworthy? Does the answer really make sense?
WolframAlpha shows how it interprets the prompt and how it comes to the result. It also provides the sources it is using. The problem is that ChatGPT not really can reason or calculate on the language model. When you for example ask about the first date in the coming years it says:
“My knowledge is based on information that was available to me at the time of my training, which ended in 2021. I do not have the ability to browse the internet or access updated information. Is there anything else I can help you with?”
However, you don’t need any knowledge to infer something like time. At the Reddit ChatGPT community, you can find many funny and interesting examples of this. Here are some of these examples:
Though the above chats are somewhat funny, it undermines the trustworthiness of the application and with that the usefulness of the answers. Still, because the conversation is much more like we are talking as humans with each other, the adoption will likely go much faster than for example WolframAlpha that is focused on math and science.
Levels
Below are several levels of knowledge:
Based on the levels you can see that current AI efforts like OpenAI and Deepmind have focused on the bottom group, Google on the bottom middle group and Wolfram Alpha on the top group. The top middle group (legal systems, ethics and common sense) is still more a reseach effort and not available in any application yet.
When I would create an AI, intuitively I would say, that it would be better to work top-down. Thus, start with correct math and logic. Then it should apply physical laws and scientific knowledge correctly. But in reality, it’s probably the other way around. AI success comes from the immediate usefulness in day-to-day jobs. AI will probably work it’s way up.
Conclusion
AI chatbots, like ChatGPT, are already surprisingly useful. You can generate Excel formulas, ask what to do on your vacation, make a lesson plan for teachers, translate a text, summarize a scientific text and so on. We just started to explore the possibilities of AI.
By exploring we found out its limitations, that we cannot browse the internet yet, that you can’t trust the output yet, that it provides wrong answers or ones that make no sense, that it’s still unstable and slow.
AI will not replace Google or WolframAlpha right away, but at the end they will merge together. It will be the number one tool to go to, like now mobile phones are the number one device for many things. AI chatbots will be our productivity tool, the first one you open in the morning. It seems far away, but in 2023 it will start to make sense.
