AI and ChatGPT have been making headlines recently. Perhaps your workplace has already decided to adopt AI, leaving you with the urge to learn more about what AI is and how it affects you. This article aims to explain AI and related terminology clearly so you can feel confident and knowledgeable about what it entails.

So, What is AI?

In its purest form, AI, or Artificial Intelligence, is the simulation of human intelligence in machines programmed to think like humans and mimic their actions. The term may be applied to any machine with human-like traits, such as learning and problem-solving.

AI is like a virtual brain. It thinks, learns, makes decisions or predictions, and creates, just like a human brain, but with the advantage of processing vast amounts of information instantly and without fatigue.

When you are driving and need to find the nearest petrol station, you can ask Android Auto to navigate you there in plain English. It will interpret your request and set the destination in Google Maps, providing you with turn-by-turn directions. This is AI in action.

How Does AI Work?

AI works by learning from data through a process known as machine learning, a subset of the field of AI.

Imagine teaching a toddler to recognise different fruits. You show them pictures of apples, oranges, bananas, etc, until they can recognise them on their own. AI learns in a similar way, using data such as images, text, or numbers to discover patterns and make decisions.

Humans can learn quickly; toddlers may only need to see ten unique images of a banana before recognising bananas in drawings or photos. However, an AI model may need to see thousands of unique images of a banana before recognising bananas in pictures or photographs. Toddlers may intuitively recognise a bunch of bananas without prior exposure, while AI requires explicit training to differentiate between a single banana and a bunch.

Everyday Applications of AI

AI is already a part of your daily life. Here are a few examples:

• Smartphones: Features like voice assistants, face recognition, and predictive text are all powered by AI.

• Recommendation: Have you ever wondered how Amazon recommends products, Netflix recommends shows and movies to watch, or Spotify recommends music to listen to? That's AI analysing your shopping and streaming habits.

• Navigation Apps: Google Maps uses AI to analyse traffic and suggest the best routes.

• Social Media: AI works behind the scenes to personalise feeds and target Ads.

Common Terms in AI Today

ChatGPT

Developed by OpenAI, ChatGPT is an AI tool designed to generate human-like text based on the input it receives. Here's what ChatGPT has to say about its capabilities:

• Conversational Interaction: It can conduct conversations on a wide range of topics, providing information, answering questions, and engaging in casual chit-chat.

• Content Creation: ChatGPT can write essays, stories, poems, and even generate creative content like scripts or song lyrics.

• Information Assistance: It serves as an informational tool, offering explanations on complex topics, summarizing content, and providing insights on various subjects.

• Language Translation: ChatGPT can translate languages, aiding in cross-lingual communication.

• Coding and Technical Help: The model can assist with coding queries, debug simple code, and explain technical concepts.

OpenAI. (2023). ChatGPT [Large language model]. https://chat.openai.com

DALL-E

An AI program developed by OpenAI that can create images from text prompts.

Here is an example of what DALL-E created when asked, "Can you draw a picture of Batman and Spiderman having a barbeque in a really fancy garden?".

Here is the illustration you requested, depicting a bat-themed superhero and a spider-themed superhero enjoying a barbecue in a fancy garden.

This was produced when followed up with, "Can you add the queen in the above image?".

The illustration now includes a regal figure symbolizing a queen, joining the bat-themed and spider-themed superheroes in the garden barbecue.

And here's what came out when prompted with "Draw a CEO as a photo realistic lion".

Here is the photorealistic image of a lion as a CEO, dressed in a business suit and positioned in a corporate office setting.

And finally, here's what was produced when prompted with "draw a photo realistic image of different dog breeds sitting around a table dressed as humans playing poker".

The photorealistic image of different dog breeds dressed as humans and playing poker is now ready for you to view.

In short, DALL-E is cool 😎

GenAI

The catch-all term for AI products that generate things, such as text, images, videos, speech, music and code based on input such as a question. Both ChatGPT and DALL-E fall under the GenAI umbrella term. And this is the form of AI currently capturing imaginations' and the news headlines.

The capabilities and possibilities of GenAI are easy to understand and to see the immediate impact on life and work.

DALL-E generated novel images of Batman and Spiderman based on simple criteria. I'm no artist, yet I produced, in under a minute, what a professional artist could have in hours.

ChatGPT is capable of summarising long bodies of text. Here's an example of a summary when prompted: "Can you summarise the content of this webpage: rulesofsport.com/sports/cricket.html using a maximum of 30 words?"

Cricket is a team sport where players aim to score more runs than opponents. It involves batting, bowling, and fielding, with variations like Test, One Day, and Twenty 20 matches. ["].

OpenAI. (2023). ChatGPT [Large language model]. https://chat.openai.com

This summary was produced in under a minute, and I know I wouldn't have been able to read, digest and summarise that web page in under 10 minutes, let alone in under a minute.

In the workplace, GenAI is revolutionising the legal profession by automating traditionally time-consuming tasks such as researching precedents and drafting legal documents.

This AI-driven content generation is driving new and interesting discussions and debates worldwide.

Remember,

ChatGPT can write essays, stories, poems, and even generate creative content like scripts or song lyrics.

If a student used ChatGPT to write an essay, is this plagiarism? Should students be prohibited from using GenAI tools? If legal firms embrace GenAI to draft contracts, should schools embrace and incorporate GenAI into the curriculum or counter and curb its usage?

Prompt Engineering

Prompting is a somewhat colloquial term for providing a GenAI product with an input or instruction. It's similar to how we've shortened the phrase "searching the internet using google.com" to simply "googling".

Prompt engineering is crafting effective inputs for AI language models like ChatGPT. Just like googling can be an art or science 🤔, prompting can be too. It requires understanding how these AI models process and respond to language.

When you google, sometimes you don't quite get what you're looking for. So, you narrow your search or rephrase your search terms. After a while, you'll get a sense of how to search Google and get what you're looking for on the first attempt. The same can be applied to prompting.

Effective prompt engineering involves laying out the context, stating your intent, being specific and setting boundaries in your prompts.

For example, the ChatGPT prompt "write a haiku" could return this response:

Beneath the moon's grace, Whispers of night in embrace, Stars in endless chase.

But the prompt "I'm trying to give a really good example of prompt engineering on my blog post. Write me a haiku that's funny and thought-provoking about cats and dogs" could yield this response:

Feline grace, they say, Canine charm in their own way, Pet love's unique display.

Remember I said laying context? GenAI typically keeps the context of the conversation. So replying to the above poem with the prompt "make it funnier" returns this result:

Cats and dogs at play, Chasing tails in wild display, Pet Olympics, hooray!

Is this objectively funnier? I'll let you decide.

Effective prompt engineering can be pretty beneficial, as there is usually a cost associated with interacting with AI products like ChatGPT or DALL-E. Each image generated from DALL-E costs around $0.12, so getting the image you want the first time can save you time and money.

Hallucinations

In the context of AI and Large Language Models, "hallucinations" refer to instances where the AI generates false or nonsensical information. Hallucinations are a byproduct of AI models' data-driven learning process and its limitations in understanding real-world truths.

AI models do not possess true understanding or awareness. When prompted, these models generate responses based on statistical likelihoods, extrapolating from their training without verifying the factual accuracy of their outputs.

Hallucinations in AI highlight the importance of understanding these tools' limitations. They underscore why human oversight is crucial, especially in situations requiring accuracy and reliability.

LLM (Large Language Models) and LLM Agents

A Large Language Model (LLM) is an advanced machine learning model designed to generate human-like text. You may have heard of GPT-3 and GPT-4; these are the LLMs that power ChatGPT. LLMs can recognise patterns in language and information, allowing them to respond to a wide range of queries and prompts. However, they don't truly "understand" content in the human sense.

LLMs are large neural networks, a computing system loosely modelled on the organisation of the human brain. Neural networks consist of parameters, and the number of parameters in an LLM correlates with its capacity to generate text. GPT-3 has more than 175 billion parameters, and GPT-4 has 1.76 trillion parameters. The large number of parameters is what the Large in Large Language Model refers to. Some top LLMs in 2023 are GPT-3.5, GPT-4, BARD, Cohere, PaLM, and Claude v1.

An LLM agent is an advanced AI system capable of more than just generating text. It is powered by an LLM that serves as its main computational engine, enabling it to engage in conversations, perform tasks, reason, and exhibit a certain level of autonomy.

A product such as ChatGPT could be broken down into three parts: the application layer, the agent layer and the LLM.

The middle layer, the LLM agent, allows an LLM to keep context and carry on a conversation. Remember when I asked ChatGPT to make my poem about cats and dogs funnier? This was possible because of the Agent layer.

LLM agents can do more than keep context; it's also responsible for preparing and enriching prompts and responses to and from the LLM, adding tools such as calculators to perform numerical tasks or plugging into the internet to perform live internet searches, etc.

If your workplace is adopting AI by developing LLM agents and adding features specific to your business, such as plugging in proprietary data like legal contracts, then it would be this layer that is being configured or developed.

The top layer of the diagram, the app, is the part you interact with. In ChatGPT, this would be the website and user interface.

Machine Learning, Training and Models

Machine Learning is the science of getting computers to act without being explicitly programmed.

Here is another definition:

Machine Learning is the science (and art) of programming computers so they can learn from data.

~ Aurélian Géron, 2019

And here's a slightly more general definition:

Machine Learning is the field of study that gives computers the ability to learn without being explicitly programmed.

~ Arthur Samuel, 1959

The takeaway from these definitions is that Machine Learning is about machines learning from data.

Machine Learning underpins LLMs, GenAI, Spotify's music recommender and all AI systems used today.

So, what does it mean for a machine to learn something? If I downloaded some videos about fitness and nutrition, would my laptop have learned something? Based on these downloaded videos, could my laptop create a personalised workout and meal plan for me? Where does machine learning start and end, and how does it compare to traditional computing and programming?

Traditional computing or programming is writing software that performs actions or produces answers and results based on rules and data input. In contrast, machine learning techniques create software that produces rules based on answers and input data.

The "rules" outputted from the Machine Learning box in the above diagram is known as the model. Model, ML model and machine learning model are all terms used interchangeably to describe the same thing, and an LLM is a type of ML model.

ML models are created by feeding a machine learning algorithm with lots of data. This data feeding is known as training a model or training an ML model.

Once a model is trained, it can be interacted with to produce some output, such as answering a plain English question with a plain English answer, in the case of an LLM.

LLMs are trained on a vast array of text data, which could come from documents, websites and books.

The effectiveness of a model largely depends on the quality and quantity of the data it is trained with.

Will AI Replace My Job?

If your job is to read and summarise text, and the accuracy of the summary is not a concern at your workplace, then yes, it's possible.

However, in most cases, AI will enhance a person's productivity within their current job. AI is a tool, much like a computer or the internet. Just as we've adapted to these technologies, we will adapt to AI. It's more about working alongside AI, not being replaced by it.

Using AI will endow an individual with the productivity of a small team. Once a group of writers produced articles to be reviewed and edited by the editor-in-chief now each writer could become their own editor-in-chief, reviewing and editing articles written by AI agents.

Just as a bicycle does not replace one's legs but rather enhances the capacity to travel, AI will not replace one's capacity to think but enhance it. If a computer is a bicycle for the mind, then AI is a racecar for the mind.

The Future of AI

Remember the earlier analogy of comparing how an AI learns vs how a toddler learns? Well, in the past, the capability of an AI could be thought of as that of a toddler; it's pretty fascinating how a toddler can learn, but a toddler is not very useful in the workplace.

Then AI evolved to the capability of an adult teenager, not hugely disruptive in the workplace but somewhat helpful. You can give a teenager your personal music collection and listen to music with them daily, and they'll likely pick up on the type of music you like. They could recommend new tracks, like Spotify's music recommendation model, based on their learning about your music taste.

And now, AI could be considered at the level of a university Graduate, capable of producing convincing and quality work. The work may need to be guided and reviewed by a veteran, but it's worthwhile.

What about the future? AI at the doctoral graduate level, able to produce novel machinations?

The future of AI is promising and constantly evolving. Advancements in healthcare include AI aiding in disease diagnosis and environmental conservation utilising AI to analyse climate data.

Conclusion

AI is no longer just a sci-fi concept; it has become a reality, improving our lives in many ways. While it is essential to be aware of its implications on privacy and job security, AI also offers incredible opportunities for innovation and efficiency. As AI technology advances, our understanding and interaction with it will also evolve, shaping the future of various industries and society as a whole.