Generative AI: Answers to frequently asked questions

Not long ago, a disclaimer like that would have seemed silly, but it’s not so far-fetched today. Recent advances in generative artificial intelligence (GenAI) show that a computer could produce texts like these. But let’s not get ahead of ourselves: You can’t ask ChatGPT a simple question and expect a well-organized, long-form article with just the right tone and message.

One thing artificial intelligence is most definitely good at is getting humans to ask questions – from basic considerations like “What is GenAI?” to existential examinations like “Is GenAI a threat to life on Earth?”

It’s a lot for the neural network between our ears to take in. We could fill several books with an analysis of just the latest advancements. At DHL Group, we are exploring ways to use AI technology to enhance our business processes. So, we thought we’d share some of the questions about artificial intelligence and GenAI that keep coming up.

But first, some numbers:

Despite our disclaimer above, we thought it would be fun to ask ChatGPT to “describe what GenAI is in 4-5 sentences.” The response is quite revealing:

As of my last knowledge update in September 2021, “GenAI” doesn’t refer to any well-known concept or term in the field of artificial intelligence or technology. It’s possible that there have been developments or new terms coined since then. If “GenAI” has emerged after September 2021, I wouldn’t have information about it. Please provide more context or details if available.

The first thing we must understand is that GenAI systems are not all-knowing, as the above response makes clear. For example, ChatGPT can’t explain any concepts that evolved after September 2021. However, change “GenAI” to “generative AI”, and you get an accurate answer. The abbreviation, which has only recently become widespread, was enough to stump the world’s most well-known chatbot.

Now to our question: Well-known consulting company McKinsey defines GenAI as follows: “Generative artificial intelligence (AI) describes algorithms (such as ChatGPT) that can be used to create new content, including audio, code, images, text, simulations, and videos.” If you want to dive deeper into machine learning and artificial intelligence, read McKinsey’s full article.

The short answer is no. Milestones in AI date back decades, but that hasn’t stopped lots of people from talking this moment up as such. Maybe those movie lovers among us just can’t help thinking of the infamous Skynet in Terminator or when Morpheus explained the “birth of AI” to Neo in The Matrix (you know, right before he pulled out the red and blue pills).

Alan Turing, often considered the father of modern computer science, speculated about machines matching human intelligence in the 1950s. Today, using voice commands to ask Alexa to turn up the volume is a normal part of daily life for many people. AI assistants like her have lived with us for nearly a decade.

Thanks to ever-increasing computing power, we can now use AI to unlock the power of big data. We’ve witnessed this firsthand in the logistics sector. Large-scale analysis of massive data volumes by AI tools has revealed patterns that have led to advances in route planning, warehouse efficiency, demand forecasting, and streamlined inventory management.

Not likely. AI might make traditional software seem old-fashioned, but use cases for software aren’t going anywhere. In fact, the trend points to an increase in conventional software solutions because AI-assisted coding has made them quicker to develop. A survey by Sourcegraph – a company that creates tools for indexing large code databases – found that 95% of the software developers and engineers polled report using AI tools to help them write code.

Traditional software will also benefit from AI’s defining feature: The ability to learn new tricks on its own by analyzing data, detecting patterns, and making predictions or decisions. This is the essence of machine learning. We won’t go into the various machine learning models, but the process of teaching a system to label emails ‘spam’ or ‘not spam’ provides an excellent example of why AI will make traditional software better. Guided by algorithms and drawing a dataset of millions of emails with both labels, a machine learning model can identify patterns that typify spam messages and calibrate a mail program’s filter accordingly. Every time the odd spam email trickles through the filter, and a user manually flags it, the AI gets new data, which it incorporates into the model and adjusts the filter. So, this machine learning model is continuously improving your mail program.

This is an important question to ask about artificial intelligence and machine learning. A lot of advanced tech gets labeled AI, but there are vast differences between a warehouse robot with machine learning capabilities and a generative AI system.

A good example is Stretch, our artificially intelligent box mover, made by our partners at Boston Dynamics. Its primary function is unloading trucks or containers, picking up boxes, and moving them elsewhere. Stretch uses several actions to complete this task: various grips, arm movements, etc. And it does all this while interpreting visual input – kind of like a human. Before Stretch saw a warehouse, it tested random combinations of actions in pursuit of a ‘reward’ (the positive outcome of moving a box). A dropped package was ‘penalized’ (recorded as a negative outcome to be avoided). Unnecessary movements were also penalized.

In other words, Stretch learned by trial and error – again, kind of like a human. The main difference is that we use our innate intelligence to figure things out. AI is the use of a robot ‘brain’ (algorithm) to figure out how to get the rewards and avoid penalties. Now, when we present Stretch with a wall of boxes, it knows which one to pick up first. And when we put Stretch in a new warehouse, the robot adapts to the unique situation.

Generative AI – as the name implies – generates something new. Ask ChatGPT to write a poem about a friendly gray cat wearing a yellow and red uniform delivering parcels on a tricycle, and it will do just that. How? It learned by analyzing a massive language dataset (called a large language model, or LLM).

The exact figures haven’t been made public, but ChatGPT’s LLM probably has some 45 terabytes of text. One terabyte is roughly 500 billion words (500,000,000,000) – far more text than any human could ever read and retain. The data is unlabeled, but by observing and categorizing specific linguistic patterns and recurring elements, ChatGPT can recreate strings of words that humans recognize as poetry.

Many experts think it won’t be long before we answer this question about artificial intelligence. For example, Gartner predicts that by 2026 “100 million humans will engage robocolleagues (synthetic virtual colleagues) to contribute to enterprise work”.

Our data scientists certainly see lots of potential, and they’re working with some of tech’s biggest names to provide the building blocks for customized solutions that we can deploy throughout DHL Group.

But at the same time, many of our colleagues worldwide are already tinkering with GenAI tools at work. We have a community of AI enthusiasts actively chatting and sharing experiences on our global employee communication platform. And it’s not just a bunch of IT techs talking to each other. Since GenAI entered the scene, more people have joined the conversation. Questions about artificial intelligence are no longer just for geeks. Anyone can get involved and explore potential ways to use the technology.

The short answer to this question is that generative AI can take over certain tedious or repetitive tasks and free us up to address more challenging and complex issues.

The National Bureau of Economic Research, an economic think tank in the US, studied more than 5,000 customer support agents and found that productivity (measured in issues resolved per hour) increased by 14% on average. You can read all about it in their April 2023 paper, Generative AI at Work.

The researchers also found “suggestive evidence” that the AI model helps less experienced workers. Newer agents who had access to AI assistance performed as well or better than colleagues with more experience but no access to AI assistance.

A survey of GitHub developers using its generative AI Copilot indicated that 87% of respondents spent less mental effort on repetitive tasks, 74% focused on more satisfying work, and 88% felt more productive.

Run the numbers and it’s easy to envision the potential productivity and efficiency gains.

Many companies are asking questions about generative artificial intelligence. And because external tools may not comply with company data protection and security policies, they’re looking for ways to leverage GenAI to create internal tools to increase general productivity across their organizations.

Whether or not your company develops its own tool or you’re allowed to use publicly available tools like ChatGPT, one fundamental question GenAI can answer is: “Where do I start?” or “What should I write?” In other words, it can help you avoid blank page syndrome (writer’s block). You can use it to create a first draft of something – like a meeting agenda or an outline for a presentation. You can use it to categorize notes, summarize a text, and draft an email. The list goes on and on. But remember that you’ll need to do some fine-tuning.

Not quite. But you should be exploring your capabilities. At DHL Group, our global Data & Analytics team is working to unlock the potential of data science. Generative AI is changing the game, turning digital solutions that were highly capable but relatively inaccessible and niche into solutions that are both highly capable and easily accessible.

Many people are asking important questions about generative AI technology, especially in light of recent advances. In late May 2023, more than 350 AI scientists and other notable figures (e.g., Sam Altman, CEO of OpenAI, and Bill Gates, co-founder of Microsoft) signed this one-line Statement on AI Risk released by the nonprofit Center for AI Safety: “Mitigating the risk of extinction from AI should be a global priority alongside other social-scale risks such as pandemics and nuclear war.”

The extinction scenario suggests that if an uncontrolled ‘artificial general intelligence’ that far surpasses human intelligence were to emerge, it could turn against humans and destroy humanity (yes, like Skynet in the Terminator movies). However, the prospect of artificial general intelligence creates as many optimists as it does pessimists: AI could also slow climate change and help cure diseases.

Could a company ‘teach’ the public large language models that its competitor is an unreliable service provider? Are business executives at risk of being ‘deep faked’ through text, photos, videos, or audio that appear to be legitimate but are, in fact, false representations generated by an AI tool? Do large language models pick up biased or unfair behaviors learned from humans?

There is also the issue of accountability and transparency, known as AI’s black box problem. Even the experts don’t fully understand what’s happening under the hood of advanced AI tools. An LLM contains an enormous dataset, but what exactly is in it? Is all information accurate? Is it ethical? Stretch the warehouse robot will only ever move boxes, but AI tools with full command of language, for example, are capable of unpredictable behavior, which can carry unintended consequences. Navigating these risks creates a much higher need for sound legal, operational, and even ethical guidance policies.

The world of generative AI is rapidly evolving, raising many intriguing questions. There are both opportunities and risks. While our disclaimer at the beginning is a tongue-in-cheek reminder that a human wrote this article, recent advances in GenAI have indeed blurred the lines between human and machine-generated content. However, it’s imperative to recognize that GenAI tools like ChatGPT still have limitations and cannot fully replace human creativity and understanding.

The potential of generative AI in the workplace is vast, promising productivity gains, improved customer support, and creative assistance – not to mention automated repetitive tasks, freeing up human resources for more complex challenges. However, GenAI systems are not a universal solution.

Companies like ours are actively exploring ways to harness GenAI to develop internal tools tailored to our unique needs. However, the world must address fundamental questions and concerns as we tread further into an AI-supported future. A collaborative and thoughtful approach will be crucial because the questions will continue.