Whilst the world jizzed over ChatGPT, OpenAI's AI chatbot, some of us had been using everything from Otter to Gmail for a while. ChatGPT seems almost magical in its ability to provide answers to a wide range of questions. But AI is way more than just chatbots – it's a diverse landscape with immense potential. While having AI-powered chatbots help you with tasks or generate captivating images is exciting (think of the fingers), the true power of AI goes beyond, potentially reshaping economies on a massive scale. McKinsey Global Institute estimates this potential impact at a staggering $4.4 trillion annually for the global economy.
How should you get started with AI? Why would you even bother? Are the old ways really the best? No. Well, sometimes yeah. But mostly ChatGPT and all its happy money-generating children are there to make you more productive. I can feel through these words I write, that you already feel more productive. Amirite? Here’s some cool ideas for ya.
Low Hanging Fruits
Want to get started quickly? These bots will help you to do stuff if you ask the right questions. I’ve added a couple to get you started.
Customer Support Chatbots: Companies use AI-powered chatbots to provide instant assistance to customers, answering common queries and guiding them through troubleshooting steps.
“Provide one paragraph about our return policy [pasted here] using a fun yet informative tone.”
Content Generation: Journalists and content creators leverage AI to draft articles, reports, and summaries, saving time and generating content ideas. “I need to write an article about AI covering chatbots and generative AI, please provide it in the style of Kelly Vero.”
The Segmenters (these were not selected for Harry Potter villains, sadly)
I loved working with DICOM and supercomputers, but unless you know what to look for, you won’t be able to tell AI what it is you need. That said, segmenting is one of THE most relevant AI processes because, like ripping a piece of paper strip by strip you want a computer that can put it back together and tell you what the data says.
Meanwhile in video games, we were creating autonomous everything over 10 years ago. The state machine that segments action to predict everything from collision to drive-bys? Yeah you can thank Forza and GTA for this revolution in self-driving cars.
Medical Imaging Diagnosis: Deep learning models analyse medical images like X-rays and MRIs to assist doctors in detecting diseases at an early stage, enhancing patient care.
Autonomous Vehicles: Deep learning algorithms process real-time data from cameras and sensors in self-driving cars to identify objects, pedestrians, and road signs, ensuring safe navigation.
Is there anything better than not thinking? Thinking makes overthinking happen and that’s no good if you want to measure and monitor. We just want the facts. So, sit back, no lay back and let AI be your Sigmund Freud. Tell it about your mother if you want. Tell it anything. Here are some ways it could help you.
Social Media Monitoring: NLP tools monitor social media platforms to gauge public sentiment about products, services, or events, helping companies make informed decisions.
“I want to know how many Gen Z users were on Instagram during September 2019.”
Customer Feedback Analysis: Businesses use sentiment analysis to analyse customer reviews and feedback, gaining insights into customer satisfaction and identifying areas for improvement.
“Here are some qualitative survey responses we received during [insert your dates here] can you provide any insight into negative customer feedback.”
The Policy People
My good friends Marion Mulder, Heidi Saas and Debbie Reynolds all use data to analyse the way that humans are treated by other humans. Be it privacy policies, HR data or the sexualisation of avatars. AI is a double edged sword here because for the most part, AI has been used as a power tool. Routing out folks based on the colour of their skin, whether they are pregnant or not (or planning to be) or whether they are just a plain old gender that does not reflect the workforce is not an AI problem, it’s a people problem. Tell AI what you want to do and it will do it. AI is the Ron Burgundy of code. It reads whatever is on the screen.
Fair Hiring Practices: AI algorithms are used to screen job applications, ensuring fair evaluation and minimising bias based on gender, ethnicity, or other factors.
Content Moderation: AI tools identify and filter out inappropriate content on online platforms, maintaining a safe and respectful environment.
Generative AI in my experience of running NAK3D is mixed. I don’t think it’s the greatest thing that ever happened to fashion. I would argue that it’s more of a segmenter. Thing is, generative AI is a massive time-saver for creatives who already get paid a lot of money, to get paid a lot of money for doing less than they did before. I’m just sayin’.
Fashion Design: AI generates new clothing designs by remixing existing designs and incorporating diverse style elements, inspiring fashion designers with fresh ideas.
Architecture Visualisation: Architects use AI to create realistic 3D visualisations of buildings based on textual descriptions, aiding in presenting design concepts to clients.
GANs are ace. Imagine you're an art forger, and you want to create paintings that look just like famous artists' works. But there's a challenge: you need to fool an art detective who's really good at telling real paintings from fake ones. Enter the GAN (generative adversarial network) which is constantly creating from data that it is constantly trained from. It’s a big step up from Kim Kardasian’s airbrushed fails, Kim I hope you are reading this.
Artistic Creations: GANs transform photographs into artwork in the style of famous painters, enabling individuals to see their photos in a new artistic light.
Interior Design: GANs help interior designers envision room transformations by applying the aesthetic features of different styles to existing spaces.
The Talkers (and the Listeners)
I mentioned Otter at the top of the article. To be honest I find it an essential tool in recording everything so I don’t forget it. But it’s good at translating verbatim too! However, the whole language and communication thing is interesting. I live in a country where English is not the national language so if I go to the doctors, I need to know that I’m getting the information or diagnoses relayed back to me correctly before I write my will. Know what I mean? I use Deepl and it’s getting better at knowing what I need German, French, Italian and Romansch skills for (mostly to get out of jail or to not die).
Language Translation: End-to-end learning models translate text from one language to another in a single step, allowing for faster and more accurate language translation services.
Speech Recognition: AI systems transcribe spoken language into written text, aiding in tasks like transcription services and voice assistants.
The Hands-off Robots (are Humans)
I am obsessed with robots. From the ones that make cars to the ones that perform surgery. The NHS is very open about wanting to use AI to assist in cutting down waiting lists and enhancing medical care performance.
Autonomous Robots: Engineers implement AI safety protocols to ensure that robots operating in public spaces make safe decisions and avoid accidents.
Medical Decision Support: AI models provide medical recommendations while adhering to ethical considerations, assisting doctors in making informed treatment choices.
Ever had a conversation with someone or something online and you’re not sure whether it’s bot or human? The Nigerian Prince has evolved! The sophistication of neural network analysis is making fraud sweet, but detection even sweeter. The linguistics of fraud, compliance and protection will continue to spill into everything from medtech to insurtech.
Financial Fraud Detection: Neural networks analyse transaction data to detect unusual patterns and identify potential cases of fraudulent activity in real time.
Language Translation: Neural networks understand the context of entire sentences and accurately translate idiomatic expressions and nuances between languages.
The Minority Report
When someone says Minority Report it’s always tinged with fear and horror but not all elements of that movie were scary. Some of it was really prescient, in ways that we’re only just discovering now in 2023. In an age where police operations are simple issuers of crime numbers, AI could really help in finding even the tiniest details. For market research, which is a totally different vertical, these details are represented by insights that span consumer behaviours. That’s an essential skill in understanding the nuances of being human.
Video Content Analysis: Multimodal AI processes both visual and audio elements in videos, enabling video platforms to provide automated subtitles and relevant content recommendations.
Market Research: Multimodal AI combines text and image data to analyse consumer reviews and photos, yielding comprehensive insights into product preferences.
The Words That Matter
As AI becomes increasingly integrated into our lives, new terms are surfacing. Whether you want to impress in conversations or interviews, here's a handy guide to some essential AI terminology:
Artificial General Intelligence (AGI): Imagine AI that surpasses current capabilities and excels in tasks far better than humans. AGI not only performs tasks but also enhances its own abilities through learning and advancement.
AI Ethics: A set of principles aimed at ensuring AI doesn't harm humans. This involves decisions about how AI systems gather data and address biases.
AI Safety: This interdisciplinary field focuses on the potential long-term consequences of AI development, particularly the risk of AI rapidly progressing to a level of superintelligence that could pose threats to humanity.
Algorithm: Think of an algorithm as a series of instructions that enables a computer program to learn from and analyse data in a specific way, like recognizing patterns, allowing it to accomplish tasks autonomously.
Alignment: Fine-tuning an AI to produce desired outcomes, from moderating content to ensuring positive human interactions.
Anthropomorphism: The tendency to assign human-like qualities to non-human entities. In AI, this can lead to overestimating the human-like attributes of AI systems (or writing about them, in my case).
Artificial Intelligence (AI): The application of technology to simulate human intelligence, either through computer programs or robotics. It aims to build systems capable of performing human tasks.
Bias: In the context of large language models, errors arising from biased training data, potentially leading to incorrect attributions based on stereotypes (Marion Mulder talks about this in LGBT+ a lot)
Chatbot: A program that communicates with humans through text, simulating human language and interactions.
Cognitive Computing: Another term for artificial intelligence.
Data Augmentation: Enhancing AI training by remixing existing data or introducing more diverse data.
Deep Learning: A subset of machine learning where artificial neural networks, inspired by the human brain, analyse complex patterns in text, images, and sound.
Emergent Behavior: Unintended abilities displayed by an AI model.
End-to-End Learning (E2E): A deep learning approach where a model is taught to complete a task holistically, considering inputs and generating solutions in a single step.
Ethical Considerations: Awareness of ethical implications related to AI, including privacy, fairness, data usage, and safety concerns.
Foom (Fast Takeoff): The notion that the development of AGI might lead to rapid progress that could be difficult to control, potentially endangering humanity.
Generative Adversarial Networks (GANs): A model composed of two neural networks – a generator and a discriminator – collaborating to create new data.
Generative AI: Technology that uses AI to produce content, such as text, video, code, or images, by recognizing patterns in training data and generating novel responses.
Google Bard: A Google AI chatbot that fetches information from the current web, unlike ChatGPT, which is limited to data until 2021.
Guardrails: Policies that guide responsible AI behaviour and prevent undesirable outputs.
Hallucination: Incorrect AI responses, resembling confident but inaccurate parroting.
Large Language Model (LLM): An AI model trained on vast text data to understand and generate human-like language.
Machine Learning (ML): A component of AI enabling computers to learn and predict outcomes without explicit programming, often used with training datasets.
Microsoft Bing: Microsoft's search engine utilising AI-powered search results, akin to Google Bard.
Multimodal AI: AI capable of processing various inputs like text, images, videos, and speech.
Natural Language Processing (NLP): A branch of AI using machine learning to enable computers to understand human language.
Neural Network: A computational model mimicking the brain's structure, designed to recognize patterns in data.
Overfitting: A machine learning error where a model performs well on training data but struggles with new data due to excessive focus on specifics.
Parameters: Numerical values shaping an LLM's behaviour and predictions.
Prompt Chaining: AI using past interactions to influence future responses.
Stochastic Parrot: An analogy illustrating that LLMs mimic human language without understanding its meaning, similar to a parrot repeating words.
Style Transfer: Adapting the style of one image to another, enabling an AI to recreate images in different artistic styles.
Temperature: A parameter controlling the randomness of an LLM's output.
Text-to-Image Generation: Creating images based on textual descriptions.
Training Data: Datasets used to teach AI models, encompassing text, images, code, and more.
Transformer Model: A neural network architecture learning context from relationships in data, such as sentence context.
Turing Test: A test assessing an AI's human-like behaviour. It passes if a human cannot distinguish its responses from those of another human.
Weak AI (Narrow AI): AI specialised in specific tasks, unable to expand beyond its designated capabilities.
Zero-Shot Learning: Testing an AI model's ability to perform tasks without specific training data, simulating real-world adaptation.
As you delve into the world of AI, understanding these terms will help you navigate conversations, grasp concepts, and appreciate the dynamic field's potential. Keep in mind that this glossary is continually evolving as AI advances. So keep popping back as I update it a little more.
Here’s your homework now, especially if you’re new to AI. Don’t read those silly infographics, instead, have a bash at using low-level AI yourself. Start working on your prompts, refine them as you learn more, and remember that all AI needs that human touch to make sure it’s not too formulaic or cringe in its presentation, and more importantly that the information provided can be fact checked. However, if you are writing anything fictional, the world is your lobster (not your Llama)! 🦞