The emergence of generative artificial intelligence has opened up a new world of possibilities linked to its exploitation and use cases. However, it also paints a scenario of uncertainty about its future capabilities, associated risks and possible limitations of its use.
Since the launch of ChatGPT in November 2022, generative artificial intelligence has been on everyone's lips and has already been described as one of the fastest-adopted technologies in history.
According to McKinsey, the public version of ChatGPT reached 100 million users in just two months, democratizing a technology —artificial intelligence— that has been around for more than a century.
Generative artificial intelligence should not be confused with "general artificial intelligence", a more abstract concept used to describe a type of artificial intelligence that will equal or surpass average human intelligence and that has been the subject of major controversy lately.
The rapid growth of generative AI can be explained by its easy accessibility. Before ChatGPT's launch, artificial intelligence already had a strong presence in the market. However, it was only in the hands of experts. ChatGPT changed the entire game by putting generative artificial intelligence in the hands of anyone.
For the first time, users can get value through generative artificial intelligence without having any knowledge of machine learning —a branch of artificial intelligence and the basis of generative AI—.
All this is possible because generative AI chatbots work with base models —expansive neural networks trained on large amounts of unstructured and unlabeled data—. The great strength of base models is that they can be used for a wide variety of tasks. Previous artificial intelligence models, in contrast, were limited to a single task.
However, this versatility also has its drawbacks. The great handicap of generative artificial intelligence is that its extensiveness makes it less accurate, increasing the probability of erroneous or biased results.
What is generative artificial intelligence?
Generative artificial intelligence is a branch of artificial intelligence that focuses on the creation of models and systems capable of generating new and original content. It uses multimodal machine learning (MML) techniques and neural networks to learn to mimic or create new instances from data that resemble examples provided during their training.
Generative AI relies on models that have the ability to learn the underlying structure of a dataset and then generate new examples that match that structure. Through this operation, generative AI models can generate different types of content, such as images, text, music and video.
A common approach in generative AI is the use of generative adversarial neural networks (GANs). GANs consist of two neural networks: a generator and a discriminator. The generator creates new data instances and the discriminator evaluates the authenticity of those instances. Both networks are trained simultaneously, with the generator trying to fool the discriminator and the discriminator improving its ability to distinguish between generated and real instances.
How is generative artificial intelligence different from other types of AI?
The main difference between generative artificial intelligence and other forms of artificial intelligence is its ability to generate new content in unstructured format —text, images, etc.—.
Generative artificial intelligence works through a type of artificial neural networks called foundational models that are trained by deep learning —a branch of machine learning—. Again, deep learning has been around since the early 2000s. However, the new functional models applied in generative AI contain certain differences from previous models.
The most significant difference of the new models is that they can be trained on extremely large and varied unstructured data sets, whereas previous deep learning models were trained on smaller and/or more specific data sets.
This, in turn, implies that previous models had a limited extent of task resolution. To put it simply, previous deep learning models could, for example, classify objects in a photograph or make a prediction using the information in the photograph. However, the new generative AI models can do both at the same time and, in addition, generate new content.
The particularity of the new deep learning models used in generative artificial intelligence is that they accumulate capabilities, generating new patterns and relationships in the data from the large datasets used for training. This is what enables ChatGPT to answer any question or generate original content and DALL-E 2 and Stable Diffusion to create new images from a description.
The versatility of generative AI opens the door to a large number of use cases that previous deep learning models could not address.
How can a business use generative AI?
Generative artificial intelligence can be used for a wide variety of things. In the business environment, its most significant use cases are based on work automation and acceleration.
Generative artificial intelligence goes beyond the capabilities of ChatGPT and can play multiple roles in an enterprise, thanks to its ability to classify, edit, summarize, answer questions and write new content.
These capabilities have great potential in terms of generating business value, as they can transform the way all areas and business processes work.
On the other hand, we must not forget that new models of generative AI are still under development and that, therefore, as this technology evolves, the use cases at the enterprise level will also increase. For now, its most relevant applications are the automation of tasks, workflows and specific actions or requests.
Nor should we forget that the versatility of this type of AI also multiplies the risks associated with its use. Therefore, it is extremely important that companies that want to adopt this technology do so while ensuring its responsible use and implementing measures to mitigate the associated risks.
The risks associated with generative artificial intelligence: Towards a responsible use of generative AI
Generative artificial intelligence presents a number of risks that CEOs need to address. It is important that they design their teams and processes in a way that mitigates these risks from the outset. This does not only involve complying with evolving regulatory requirements, but also protecting the business and earning the digital trust of consumers.
Below we review the risks associated with generative AI that must be taken into account for its exploitation.
- Impartiality: Generative AI models are prone to generate algorithmic biases due to imperfect training data or poor decisions made by the engineers developing the models.
- Intellectual property (IP): Training data and the results of generative AI models can generate significant intellectual property risks, including infringement of copyrights, trademarks, patents or other legally protected materials. Even when using a vendor's generative AI tool, companies should analyze what data has been used to train the algorithm and how it is used in generating the tool's results.
- Privacy: Generative AI can raise privacy concerns if users enter information that later appears in the model results in a way that allows individuals to be identified. On the other hand, generative AI can also be used to create and disseminate malicious content —misinformation, deepfakes and hate speech—.
- Security: In the wrong hands, generative AI can be used to accelerate the sophistication and speed of cyberattacks. It can also be manipulated to provide malicious results. For example, through a technique called instruction injection, a third party gives a model new instructions that trick the model into providing an outcome not intended by the model producer and end user.
- Reliability: Generative AI models may give different answers to the same questions, which prevents the user from assessing the accuracy and reliability of the results.
- Social and environmental impact: The development and training of foundational models can have detrimental social and environmental consequences, including increased carbon emissions.
The democratization of generative artificial intelligence places us in an unpredictable scenario that raises many questions, challenges and opportunities.
Companies have been striving for years to exploit the potential of artificial intelligence. The sudden emergence of generative AI does not mean that the exploitation of old AI models has become obsolete. However, generative AI represents an exciting and unknown leap forward that opens up a world of new possibilities that will continue to expand in the coming years.
In this context, it is essential that the development and use of generative AI progresses at a pace similar to that of its policies, regulations and risk mitigation assumptions.