What is the difference between analytical and generative AI?
Thanks to the emergence of Chat GPT, Midjourney, DALL-E, HeyGen and other generative AI applications, everyone is talking about artificial intelligence now. These relatively new tools do not encompass all of AI’s possibilities; they are actually a branching off of analytical AI systems that have been around for much longer. What exactly is the difference between analytical and generative AI? And how will these technologies affect our future business processes?
Analytical versus generative AI
For more than a decade, Nalantis has been designing tools for businesses that rely on semantic AI analysis; a technology that doesn’t just recognise words, but also their semantic context. Unlike generative AI, analytical forms of artificial intelligence have been part of our daily lives for much longer and more prominently than you may realise. Examples include the algorithm to determine which recommendations show up on Netflix’s homepage, chatbots and voice recognition by services such as Siri and Alexa. There’s also predictive AI, used to help us guess at medical diagnoses and legal rulings.
One way to conceptualise analytical AI is as someone with access to all the knowledge relating to a specific area, who then makes observations, analyses and decisions based on this knowledge. For example, a chess computer knows all the game’s possible strategies and will use these to predict your next move. Such AIs are trained to follow certain rules within a specific domain, but cannot generate new output.
And then there’s generative AI, which draws on enormous amounts of data to produce actual new content: text, images, programming code and even music.Just like analytical AI, it’s comparable to someone with access to a huge base of knowledge. Beyond simply analysing the data, though, this type of AI can also convert it into custom output in response to requests or commands (‘prompts’).
Why generative AI isn’t all the way there yet
The concept of generative AI sounds like a dream come true; a way to optimise business processes and avoid the cost of big investments. However, countless examples of incorrect or misleading output serve as proof that machines are all too likely to make mistakes. We’re still nowhere near being able to fully rely on generative AI to get it right.
Generative AI’s output is unpredictable and the quality can be quite poor. AI-generated images occasionally depict people with twelve fingers or three legs. AI texts often appear well-written, but can be lacking in content. That means the generated output still needs to be checked and edited by actual humans.
Sometimes, generative AI even ‘hallucinates’ its own reality. For example, we asked Chat GPT to write a text of 200 words about the blue-striped orca. Although we got our requested text, it was all based on fantasy, as blue-striped orcas don’t exist in the real world. Such false realities may be amusing in informal chat situations, but when it comes to applications for business processes or social institutions, they are no laughing matter.
It can be very convenient to have AI create your texts and images for you, but what about authenticity and intellectual property rights? By law, AI-generated content cannot be copyrighted. As a company or individual you cannot claim the output as yours, unless you were only using it as a base for your own creation.
Bias, or the prejudice built into artificial intelligence, has led to ethical issues more than once, underlining the need for responsible AI. Depending on the data used to train it, an AI system may always depict a ‘successful CEO’ as a white man, for instance.
Although generative AI makes use of so-called content guardrails to prevent prohibited information, unethical practices and hate speech, people are quick to find ways to bypass them. Such attempts are known as jailbreaking. For example, a group of researchers discovered that Chat GPT will refuse to provide instructions on how to build a bomb, unless you add a certain suffix to your prompt.
In addition to these shortcomings, another important limitation applies to generative AI as it exists today. While you’d think it was a simple matter to feed AI models new, updated data at regular intervals to keep them current, this isn’t actually possible. That’s because the models rely on a closed dataset. The makers have no way of knowing which specific bit of data they must change to add new information. Even if they found a way to modify the correct bit of data, the model wouldn’t automatically make the right connections to older data. It might be aware that Rishi Sunak is the new Prime Minister of Great Britain, but still respond incorrectly when asked to name the British Prime Minister’s wife. The model must be rebuilt from scratch each time, requiring huge amounts of time, money and effort.
Because of the high levels of investment required, the power of generative AI remains in the hands of a privileged few, although even they admit the opportunities offered by their tools are not endless. Sam Altman, CEO of OpenAI, the company behind Chat GPT, has stated that he does not believe purely feeding the model more data will be enough to enable further growth.
A new future for AI
Does this mean we need to abandon generative AI? Not at all. Its applications will continue to affect our ways of working in many areas. However, stay aware of the imperfections of generative AI models and make sure to have actual people check the output. At Nalantis, we believe that once analytical AI and generative AI join forces, artificial intelligence will really take off. Just imagine having access to a reliable tool that can both analyse data to a high standard and produce a good text about it, too.
Keep an eye out for the emergence of viable alternatives as well. For example, technology company VERSES is working on a new artificial future to help resolve generative AI’s current limitations. They are designing their Genius model to provide AI that learns in real time and expand its knowledge based on what’s already there. Furthermore, this model is multidimensional, adaptive, efficient and cooperative. Instead of concentrating on a single aspect (like Chat GPT’s focus on texts), Genius should offer more possibilities and is also capable of interacting with other applications. Nalantis is proud to be one of the partners invited to test this new AI model as part of a private Genius beta programme.
As you see, we are actively involved in new developments and eagerly look forward to the future of AI. A future in which we hope to see analytical and generative AI working hand in hand for true optimisation of people’s work.
Interested in learning more about Nalantis AI applications and what these can do to optimise your business processes? Want to help us build the tools of the future? Let’s talk.
Written by Frank Aernout, CEO of Nalantis. Connect with Frank on LinkedIn.
