A few days ago, as I was walking to the grocery store, I verbally asked Bing ChatGPT-4: “Can you come up with a three-course menu for a scrumptious dinner for eight? Create a combined shopping list organized by grocery aisle? Merge the cooking instructions of the recipes?

Tell me what to do sequentially to have everything ready at the same time? Can you also create goofy names for the courses that I can print out as a menu?”

I had my answer four seconds later.

Welcome to a new world. In the future, we will look back at this moment in history and realize this was when everything changed.

Machine intelligence has been improving for decades, and today, nearly all of the business services we depend on have been optimized and personalized by them.

But that was only the beginning. Also today, profoundly novel Generative Artificial Intelligences are stepping into the spotlight. “I am your personal tutor and partner in all things creative,” they tell us. It is a big claim, but it turns out to be mostly true. They are utterly astonishing.

Due to the vast amount of free human knowledge accessible online, Artificial Intelligence (AI) tools have had access to enough material to train the speedy neural net learning machines we call “transformers” into the glib predictive geniuses we call Large Language Models, or Generative AI.

One of those is ChatGP-4. It absorbed trillions of items of human history, and like other Large Language Models, the results enabled uncanny abilities that are far more expansive than the sum of their parts.

GPT stands for “Generative Pretrained Transformer.” The term “generative” refers to the model’s ability to respond to a prompt.  “Pretrained” tells us that this AI model has consumed large data sets or been trained by another AI that has been fed large amounts of data that align it to its given purpose.  And “Transformer” refers to the use of a particular neural network architecture for this Model. GPT is a step toward what many developers think is the holy grail, “Artificial General Intelligence,” or AGI, as it is often called.

AI is not intended to be a replacement for human intelligence. In fact, its greatest value is that it does not think like us. It can be faster and, if carefully checked, far more logical than we will ever be. It makes for a wonderful assistant and partner, able to augment our work, and as it continues to improve, it will be as immersive and indispensable as the internet.

The benefits of interactive generators fall into three categories.

Learn: AI Chatbots can explain just about anything to people in a way they will understand.

Invent: You can partner with the new AI Chatbots to create videos, music, images, or author-entertaining written content.

Research, Deduce, Predict: By partnering with AI Chatbots, we can reveal insights that would have taken years of tedious research. They can help deduce protein folding, genetic markers, the language of mRNA, predict virus mutations, map astrophysical phenomena, oceanography, financial markets, business strategy, psychology, and climate change. As professional tools, these AIs have a way of becoming indispensable quickly.

AI is not intended to be a replacement for human intelligence. In fact, its greatest value is that it does not think like us. It can be faster and, if carefully checked, far more logical than we will ever be.

For example, AI is being used to monitor sensors that track corporate adherence to environmental laws that were previously too time-consuming to track. They then fold that data into a carbon footprint visualization of an entire industry across continents so we can determine the impact of adopting innovative manufacturing technologies and calculate when the industry will be able to meet international emission standards.

The scope of these tools is vast. They can develop step-by-step instructions for a task, illustrate an idea you are formulating, summarize a lengthy article, locate a specific scene in a movie, or predict the next evolution of a pathogen like Covid 19.

 

The Wicked Cool Nature of Threads

A Chatbot’s first response to your prompt is a starting point.

The art occurs when you further refine. You can add the intended audience by saying, “Rewrite that as if you were explaining it to an 11-year-old,” or you could say, “Can you repeat succinctly and in two sentences the same essay you just wrote but this time tailor it for a TikTok user to read?”

Each time you give Chatbots feedback, they further align and predict what you need, and you’ll want to keep those refinements in a single thread so that your Chatbot can remember how to personalize and predict the information you need the next time you work in that particular area.

For example, I currently have threads on gardening, history, psychology, microbiology, the Web telescope, and cooking. Each thread is a self-referential history of questions and answers in that subject matter, complete with sentence structure nuances and critiques from me to increase efficiency.

If I am in my gardening thread and I ask about the watering needs of a Roselle Hibiscus plant, the AI’s response includes things it predicts I will likely ask next, such as whether the plant is edible or toxic; the days to harvest; preferred soil pH; hours of sunlight needed; the best companion plants; and even a couple of recipes.

Bias, transparency, and who paid for the AI training?

If you are working on something important, only use AI Chatbots from reputable training companies and follow each prompt with a request for the source sites used to compile the answer. Ask, “What sources did you compile to create the previous response? Can you provide those links and highlight the important areas?”

Some Chatbots automatically provide you with links. Either way, you will need to double-check the sources listed as they are not always accurate.

Faulty AI responses are called hallucinations. After discovering a few of these, I found myself reviewing the source websites before even reading the response. More than once, I have caught a misquote or a source with, for example, less reliable health advice and, thus, asked the Chatbot to look at more reputable data for the answer.

The previous head of the Allen Institute for AI, Oren Etzioni, points out that there will be AI Chatbot tools with a Democratic-leaning bias and others with a Republican-leaning bias. By identifying who paid for an AI’s training, you can make a guess as to your chatbot’s disposition.

This is a serious area of concern as some AIs reinforce harmful stereotypes and perpetuate discriminatory practices due to the material they were trained on or the views of the root programmers. Unknowing users have employed these tools to help with hiring, only to realize later they were screening out 50% of qualified candidates, the women, since they had assumed from their training data that engineers were always men.

 

Are you worried about your own gullibility?

Being gullible is something we have dealt with before. History is full of snake oil salesmen, and fake news has been around for hundreds of years. We have learned not to fall for pitches from card dealers on the corner, and we have learned, after numerous costly mistakes, whom to trust or believe, just as we have become savvy about not putting our wallets or phones where they could be snapped up by a thief. We can master this new challenge if we understand the mechanics behind this duping epidemic.

For example, you have likely heard that before a woman even knows she is pregnant, marketing experts have used data mining to reveal there is a new addition in her womb. They analyzed her purchases of unscented lotion, cotton, and food. They have keenly recognized a particular cadence that identifies the first weeks of gestation and predicts an accurate due date.

Today, those same marketing experts are employing billions of dollars of computing power to study “contextual signals” to determine what will elicit an emotional response from us that they can optimize. Ultimately, they need to do one simple thing: rile us.

Two important things to keep in mind: Do not make decisions under the influence of emotions and discern the bias of your tech tools.

 

Equitable AI leads to superior AI

There is a new field of science called Algorithmic Auditing that aims to identify and address biases in AI algorithms that have the potential to lead to unfair or discriminatory outcomes that negatively impact certain groups of people. This topic is important and must be done, but the goal of equitability is not just about removing biases. It is also about creating a far greater product.

Two reliable studies, one from McKinsey & Co. and the other from the Harvard Business Review, found irrefutable proof that highly diverse teams solve problems faster, make better products and, in the case of commercial companies, increase stock value and fare better financially. The European Union has already created provisions mandating AI equitability. Let’s hope we will follow suit. If indeed this is a new global arms race, then the winning technology will surely be the one that is built by a highly diverse team.

Our present is a brief moment in this far grander story, and if we want that future to include partnering with safe AI Chatbot tools for the betterment of all, we need to speak up. Where this story will lead has not yet been written, but the wicked cool thing, as the writer Jeanette Winterson sometimes says, is that we are the ones still holding the pen.

 

 

Petra Franklin is a steward of the Seattle tech community, a prolific venture capitalist, and the cofounder of Dwehl. She is a proscience public policy thinker dedicated to the ethical issues of neuroscience, biotech, and emerging technologies. She was a founding member of the Women’s Bioethics Project, chair of the National Science Foundation’s advisory board for the Material Science Center, and worked for Sun Microsystems, PBS, and CBS. She is a regular participant in the global investment business TechStars and news site GeekWire communities.

 

Rob Stein is a traditional and digital artist of 40 years, now concentrating on the new tools of Artificial Intelligence.