In the last year, we have seen huge leaps in generative AI. Social media platforms have been abuzz with news about ChatGPT, Dall-E, and other large language models that could produce content, images, and even code based on prompts.
These solutions often produce results in a few minutes or even seconds, which may take a human hours or even days to produce. Generative AI promises to completely transform industries and how businesses and employees approach some of their tasks and activities.
That said, there are also many risks associated with generative AI. Without due diligence, organizations may find the use of generative AI too costly.
Let’s explore what generative AI is, and the impact it may have on businesses.
What is generative AI? How does an AI language model work?
Generative AI is an artificial intelligence system that can produce text, audio, images, and even video based on prompts. For instance, ChatGPT which became very popular after its recent public release, can answer general questions, write essays, provide cooking recipes, and more based on questions or prompts from the user.
Under the hood, a generative AI model uses machine learning systems or artificial neural networks to process input(the prompt or the question) and produce the output. Artificial neural networks have been around for a long time, powering everything from digital assistants to self-driving cars. They mimic the human brain and are made of artificial neurons. As the name suggests, machine learning models learn from examples or training data.
For instance, imagine you’re building an artificial neural network that can distinguish between images of cats and dogs. To train the neural network, you’ll need tens of thousands of images of cats and dogs labeled as such.
Similarly, a generative AI model is also trained using vast amounts of data. For instance, ChatGPT-3 was trained on around 45TB of text data while DALL-E was trained on around 12 million images.
ChatGPT uses a language model called GPT-3 or Generative Pre-trained Transformer-3. By using large volumes of text, the model is trained to predict the next word in a sequence of words.
DALL-E uses GPT-3 to understand the prompts. Then it uses another AI called CLIP, or Contrastive Language-Image Pre-training. CLIP has been trained on a large data set of images and captions and can encode an image based on a caption.
In simple terms, CLIP works as a reference for AI language models, to connect images to text and vice versa. Now CLIP can understand the text and how an image may look like, but it can’t produce an image. For this, DALL-E uses GLIDE, another AI model.
As you can see, there’s no single approach to creating a generative AI model. Different applications require different AI models.
How can businesses use generative AI models?
Generative AI models are still in their very early stages and businesses are only beginning to explore how they may use it in their business processes. As the technology evolves and matures, we may see more exciting applications. That said, many organizations are already putting generative AI into use. Here are some of the most promising and exciting use cases for generative AI:
1. Accelerating research
Traditionally, research and development used to take up lots of resources and time. Products and solutions in all industries go through a large funnel of ideation, validation, prototyping, testing, and further iterations before they reach the market. In the pharmaceutical industry, developing a single drug may cost billions of dollars and up to 20 years of research.
With generative AI models, researchers can reduce the cost of research to a large extent. Generative AIs can produce novel products or solutions based on prompts and researchers can go straight to the validation and testing phase.
IBM is already working with researchers to accelerate drug discovery. Using the technology, they were able to find potential drug candidates to combat antibiotic resistance.
Besides generating potential solutions, generative AI may also be able to assist researchers to find relevant information from large collections of research papers and studies. Instead of manually combing through websites and trying to assimilate all of the knowledge, researchers can get a quick brief on just about any paper or study within seconds.
2. Generative AI can power advanced chatbots
Chatbots have been around for a while, but let’s face it, most of them have been very clunky and could hardly replace a human being. Unless customers used very specific phrases or keywords, chatbots struggled to understand them and created a frustrating experience for everyone.
But generative AIs have already proven their ability to create almost human-like messages. They can understand follow-up questions and provide answers just like a human being. Microsoft has already integrated ChatGPT into Bing, and Google is working hard to catch up.
With generative AI, businesses can use advanced chatbots to assist their customers. Customers don’t have to wait around for a customer support agent, and businesses can use their support teams to analyze trends and enhance the overall user experience.
3. Faster product development
Product development has been traditionally a time-consuming process. A major part of this is designing and prototyping. Even when you have an idea in your mind, you cannot really know if it is feasible until is designed and prototyped.
Every new design and iteration takes considerable resources. By leveraging generative AI tools, designers can have a product idea and have a design in minutes. Tools like DALL-E and Midjourney can create photorealistic images based on prompts within minutes. Vizcom, a powerful AI solution can create 3D illustrations out of 2D drawings.
These solutions can help drastically cut down product development timelines for businesses.
4. Marketing and sales
Marketing and sales were some of the obvious use cases for generative AI. Marketing teams regularly need tons in the form of copy and images for promotions and campaigns.
While generative AI technologies are at least decades away from replacing copywriters and graphic designers, it can certainly speed up their work.
5. Building applications
Generative AI solutions are already helping developers write code and build applications. ChatGPT can produce code for simple enough applications and Github co-pilot can help developers create applications within a very short time.
By using these generative AI solutions, businesses can build robust applications for both internal and consumer use within a short time and with few resources.
What industries will be impacted mostly by generative AI?
Since we’re in the very early days of generative AI, we cannot give a definitive answer to the question. Technology may very well impact all industries to some extent. We can expect manufacturing industries to use generative AI to come up with multipurpose tooling and we can expect generative AI to come up with innovative loans or funds in the financial sector.
In the short term though, we can expect the biggest impact on the marketing sector. The capability of ChatGPT to come up with content and copy is already shaking up marketing firms. Many of them have already incorporated ChatGPT into their workflows while many are vehemently opposed to using ChatGPT.
We can also expect changes in the education sector. Teachers and professors have expressed concerns about students using ChatGPT to write essays and assignments. As a response to this, Turnitin, among others has rolled out solutions that can detect AI-written content.
Since ChatGPT is still fairly new, we don’t know how it may change the education sector. Schools and universities may have to come up with new ways of gauging performance or AI detectors may evolve along with AI writing solutions. Educational institutions may even encourage students to use these technologies to produce creative results. Universities may even roll out courses that teach prompt engineering or other tools that can help students make the best use of these technologies.
While currently, we’re not seeing any use cases beyond research, we can expect finance and legal sectors to adopt this tech to streamline their processes. Lawyers may use the technology to draft legal documents while finance experts may use them to come up with new policies or investment strategies.
We can also expect generative AI to have serious impacts on the entertainment sector. We may see artists using generative AI to come up with new music or videos or even new forms of installations. While we are yet to see this, it may not be long before AI creates full-length movies.
What are the risks of using generative AI tools?
While generative AI can help businesses in many different ways, there are also many risks associated with using it. Once again, since the technology is very young, we are not yet sure how these issues may affect a business.
Generative AI solutions are trained on existing data, produced largely by humans. For instance, Github Co-pilot is trained on code from public repositories, and AI solutions that produce artwork are trained on artwork from human artists.
This raises many copyright questions. The first and most important of these is who is the owner of the product from the generative AI? Is it the person who created the prompt and produced the output, is it the organization that made the AI model, or is it the people who created the training data?
Another question is if the people whose work was used to train these systems will ever be compensated. For instance, Reddit has already made it clear it will charge companies that use conversations on its platform to train their language models.
While there are no laws yet that cover the output from generative AI systems, future legislation or lawsuits can threaten companies using these technologies.
Security experts have already warned against entering confidential information into ChatGPT since the communication between the user and the server isn’t encrypted.
If companies are not running generative AI tools— which use considerable computing power — within their own servers, they run the risk of data loss or theft. There’s no guarantee that the companies offering generative AI tools won’t use the data from your company to build better systems or gain a competitive advantage.
Risk of errors
Users have pointed out that it’s best to take the output from generative AI with a grain of salt. There have been numerous instances of ChatGPT producing answers that seem correct but were blatantly wrong. While the goal of ChatGPT was never accuracy, it does suggest that the technology can produce unintended results.
A generative AI that can produce possible drug candidates or designs for a new wearable may also similarly produce results that would never work in a real-life scenario.
Best practices for using generative AI in business
Verify results from generative AI solutions
This is largely about text-based generative AI solutions but it’s always a good idea to verify all results from generative AI. These systems are not error-proof; in fact, many have observed ChatGPT straight-up making up facts and sources.
Even if OpenAI or any other organization does come up with a generative AI solution promising high accuracy, it’s always best to verify the information independently.
For now, at least, limit generative AI to internal uses
We may be too cautious with this approach and this does limit generative AI to a large extent, but the copyright risks associated with it are too high for external use. There’s no legislation on how copyright may apply to generative AI solutions but it won’t be long before it will be regulated.
At that point, your company may be affected if any of the creators whose data was used decides to go for a lawsuit.
If you cannot limit to internal uses alone, you may have to do a risk assessment and limit the use accordingly.
Don’t use confidential information with generative AI systems
AI systems need a ton of data to learn from and companies that are building these solutions are always on the lookout for more of it. Unless they specifically guarantee that they’re not using customer data, they’re collecting it for training.
If you’re working with sensitive information, don’t use generative AI systems to process it.
Stay tuned with industry trends and best practices with Cadynce blog
Cadynce blog brings you the latest in automation, workflow, process management, AI, and other topics of interest for management, finance, and manufacturing sectors. At Cadynce, our goal is to help professionals stay productive and produce good results.
Stay tuned for more posts. And check out our no-code business improvement platform.