When Generative Artificial Intelligence (GAI) arrived almost overnight, it revolutionised how we create content. Everyone can now generate fresh, original multimedia content using GAI algorithms, including text, audio, video, and more.
From the outset, it’s essential to understand how GAI differs from conversational AI systems. Rather than generate content, this technology aims to identify patterns and make predictions from specific pre-defined data.
In this blog, we will take a rounded view of GAI and how it can help us while acknowledging some of its limitations.
With only a cursory news media review, it’s clear that GAI will impact almost every aspect of our lives. However, there are inherent risks and uncertainties, with some (warranted) disquiet expressed by key influencers in the tech field.
The term “generative artificial intelligence” hit the mainstream in November 2022. After many years of research by software engineers, apps, including ChatGPT appeared seemingly from nowhere. Early adopters immediately saw the fantastic things GAI could do. We now find ourselves at an AI tipping point. Finally, we have AI tools to help businesses generate a vast array of creative content or enable large corporates or governments to develop GAI-enabled products and services.
The potential of GAI is exciting, but at this stage, the casual user doesn’t fully know what these tools are capable of or how they work.
What is Generative AI?
GAI is a technology capable of independently generating original content, including audio, code, graphics, text, simulations, and videos. The technology achieves this by referencing the data on which it has been ‘trained’. For example, the popular OpenAI GPT tool, ChatGPT, is trained on a base large language model called GPT-3 that enables users to create unlimited content based on this data.
As important for this groundbreaking technology are the simple, intuitive interfaces used to request content creation that put the skills of writers, graphic designers, musicians or coders into everyone’s hands.
So, How does Generative AI work?
Central to GAI is the concept of training. This is the stock of data and information a GAI algorithm references and manipulates to create new multimedia content. The data sets of generic mass-market GAIs, like ChatGPT, Bard or MidJourney, can be extensive (GPT-3, for example, was trained on 45 terabytes of text data).
When used in a specific or regulated corporate setting like scientific or legal environments, the data generated must always comply with and follow strict regulatory guidelines. In these circumstances, the AI and its training data undergo a rigorous conditioning process, including contextual understanding, bias reduction, expert knowledge integration, quality control, and the incorporation of user feedback. This ongoing process ensures that all outputs are scrupulously accurate and compliant.
Underpinning all GAIs are Generative Adversarial Networks (GANs), a subset of deep learning technology. GANs are the product of two neural networks: a generator that renders new data and a discriminator that assesses the resulting outcomes. Working together, the generator and discriminator produce the requested content. In addition, the GANS continuously provide each other with feedback that helps to fine-tune the outputs it develops.
The other factor in promoting the originality of generative results comes from the amount and quality of data used to train a GAI. Clearly, in every use case, this exceeds the referencing capabilities of human creators by a considerable margin.
Finally, generic GAIs will have randomisation components, enabling them to generate various outputs from a single input request, giving the impression that they are genuinely ‘creative’.
Benefits of Generative AI
Even at this early stage, the potential of GAI is obvious. The range of uses for this technology is growing by the day. Notwithstanding the (already proven) ability to interpret and generate defined content, developers are investigating how generative AI can rationalise workflows. Examples of how GAI will help us in the future include:
1. Consolidating complex knowledge into a more easily digestible format
2. Automating the response to specialised technical questions
3. Content creation in any given medium
4. Acting as a co-creator for specialist designers or academics to work more effectively and with more significant impact.
The overall economic benefits of GAI are seemingly without limit. For example, GAI’s ability to work with data will eliminate repetitive, time-consuming work such as data entry, financial consolidation or document reviews. In addition, GAI frees up human bandwidth to pursue more strategic, creative and value-added work by eliminating mundane, repetitive activities.
Naturally, this technology may also displace a significant chunk of human economic activity, particularly in the creative sector and areas such as customer service or wholesale data entry.
The Limitations Of Generative AI
For a technology with so much potential, it’s important to remember that it remains an emerging technology. Even at this early stage, issues are developing, and a degree of risk remains, as highlighted in apocalyptic terms by Elon Musk:
“AI has the potential of civilisation destruction.”
Even if we take this dystopian idea with a pinch of salt, we already know mass-market GAI, like ChatGPT, Midjourney, Bard and the like, have inherent issues:
1. Sometimes the outputs generated are plain wrong. Worse, it can be disturbingly discriminative based as it is on the biases expressed online by human creators.
2. Bad actors could use GAI outputs to excuse unethical or illegal conduct, especially when taken at face value.
3. It does not always identify the origin of the content generated.
4. There are growing concerns about privacy, security and copyright contravention.
All these challenges must be addressed if GAI is to become a core part of how we live and work.
Today GAI is a largely unexplored field, and we are only beginning to understand what this technology will offer. Additional applications of this tech will emerge rapidly, with new uses likely to drop with surprising regularity. The AI snowball is only just getting going, but we’ll see GAI plays an increasing role in business, society, and our personal lives.
In parallel, as we learn and experience more of what it can do, society’s attitudes to the advantages and risks GAI poses will shift alongside how we exploit and mitigate its potential. In that regard, we can expect the emergence of new regulatory regimes and legal interventions to help us make GAI a force for good.
Contact our team to schedule a demo or to discuss a free trial of Ariya, phamax’s leading AI digital assistant designed exclusively for the healthcare sector.