Dr. Djamila Amimer is the founder of Mind Senses Global, an AI management consultancy, which helps businesses and organisations apply AI and unlock its full potential. Djamila is an experienced business leader and an entrepreneur with a broad range of experiences across AI and Business. She has a PhD in Artificial Intelligence and Energy and developed novel frameworks and AI techniques in the area of investment decisions, dealing with uncertainty in project evaluation. In addition to helping organisations reshape their business using AI, Djamila spends significant amount of my time exploring the next wave of AI. Djamila is ranked among Top Thought Leaders in AI.
ChatGPT frenzy release did not go unnoticed! You can hardly ignore the fact that it has crossed one million users within a week of its launch and sparked a $10 billion investment discussion with Microsoft. There is no doubt that 2023 is going to be an exciting time for Generative AI!
But What is Generative AI?
To put it simply, Generative AI is a subfield of Artificial Intelligence, which uses complex machine learning models to generate text, images, videos, and code. Unlike many AI subfields and applications that focus on analysing data and detecting new patterns, the main focus of Generative AI is to generate novel content.
There are two main techniques used in generative AI: generative adversarial networks and transformer based. The latter is used in large language models. Examples of Generative AI language and text-to-image models include Google LaMDA and OpenAI ChatGPT, GPT3 and DALL-E2. Generative AI models such as large language and text-to-image models require significant computing power and huge data sets to be trained on.
Most Generative AI models generate text or images outputs based on an initial prompt from the user. Users may need to combine multiple prompts or go through several iterations as the quality of the output is directly impacted by the quality of the prompt.
What are the Top Applications of Generative AI?
There are several applications and outputs that are generated by AI including:
- Generating text: users can use generative AI models to write blogs or articles. This is a useful tool for content generation. The AI models could also be used as agent/ assistant that you can interact with via dialogue to answer questions for example.
- Generating images: by using generative AI models, users can transform text into images. Prompts could be used to guide the model in terms of colours to use, the style, what to include in the image and what to exclude. This functionality is very useful for creators and designers.
- Altering images: existing images could be altered using AI models such as changing the background, the colours, and the lightings in the image. Generative AI could also be used to enhance the current resolution of images, this could be very useful when dealing with old materials for example.
- Producing videos and music: generative AI models can be used to write a script for a video, which then could be converted into a video. It can also be used to guess the next sequence in the video, this is a handy functionality if dealing with HSSE risk detection/ prediction for example. Similarly, generative AI can be used to produce music.
- Generating Speech: by using generative AI models, text such as a script, or a book can be converted into speech. This is very useful tool if we want to tap in into the audio market.
- Generating code: generative AI models could be used to produce code; it can also suggest ways to improve an existing code/ programme. This is a very useful functionality for both technical and non-technical professionals.
What are the Current Challenges and Opportunities?
Gartner’s 2022 report on Emerging Technologies and Trends Impact Rada has classed generative AI as one of the most disruptive technologies.
Generative AI is expected to have significant impacts across several sectors of the economy such as logistic, manufacturing, marketing, and healthcare. Gardner predicts that 50% of drug discovery will be driven by generative AI, thanks to speeding the process of identification of potential drugs and the testing of effectiveness. Generative AI can also be used in other areas of healthcare such as personalised medicine and medical imaging.
The main draw backs and concerns about AI are around ethics and bias and generative AI is no different. There is no doubt that if the generative AI models are trained on biased data, the generated content will be biased too. Large language models are trained on huge data sets from the web and social media, these models are prone to bias as there is no sure way to ensure the quality of data and no guarantee that the data is bias free. There have been several studies that showed that large language models contained implicit bias towards gender, ethnic groups and people with disabilities.
Another concern is copyright issue. Because generative AI models are trained on millions of data sets, creators are concerned about copyright infringement. There are also concerns whether AI generated content should compete with human generated content. In September 2022, an AI artwork won the first place at the Colorado State Fair’s fine arts competition, artists were furious that AI generated arts could enter the competition.
Other concerns include the costs and the carbon footprint of Generative AI models. The size of large language models is getting too big. For example, OpenAI’s GPT3 has 175 billion parameters, its training costs were about $12 million, and it is estimated that its associated carbon footprint was equal to driving to the moon and back. The new GPT4, which will be released sometime during 2023, is expected to have 100 trillion parameters.
Now that OpenAI made its move there is no doubt that the tech giants will enter into a battle of the titans. The battle will not only focus on conversational AI but will spill into other areas such as search engines with Microsoft plans to combine ChatGPT with Bing. With commercialisation in mind, we will see business applications extending to sectors such as marketing, advertising, legal and customer services.
However, despite the latest successes, Generative AI models lack crucial capabilities such as understanding and reasoning, addressing AI ethics and bias needs to remain a top priority.