The ripples of the current AI boom are being felt in all industries, with potentially groundbreaking benefits and fear-inducing pitfalls being touted in every corner of the media.
But amidst all this, there has been a greater tangible effect: namely a widespread desire to adopt generative AI specifically, with boards and C-suite executives questioning their organisations' readiness to seize the opportunities. At Mesh-AI, we are having more conversations than ever before with enterprises who want to embed generative AI into their business.
The quirky side-effect is that the more ‘traditional’ use cases of AI, such as natural language processing, predictive analytics and computer vision (also known as image recognition) are being forgotten despite their ability to fix real-world problems. Generative AI’s speed to adoption demonstrates its value, but it’s not the sole path to adopting transformative AI.
Mesh AI's industry experts have highlighted five major trends within AI and how enterprises can capitalise on the opportunities.
For years we've heard and seen the upwards arrow of descriptive to predictive to prescriptive. But this has only been realised in a small subset of areas. Predictive means making well defined predictions of the future based on the past. Prescriptive means turning these into actionable recommendations, creating value drivers and actionable innovation through data. New models like foundation models are able to multi-task and, when combined with the traditional prescriptive methods like recommendation algorithms, can step change applications.
The wave of newly accessible and scalable AI architectures like generative AI, combined with the likes of knowledge graphs and traditional machine learning, really do give companies the ability to go from predictive to prescriptive at pace. Combined with the continued advancement of managed software, scale of compute and ease of integration, these prescriptive solutions can be deployed at pace to impact your business in a myriad of new value driving ways, be that financial forecasts, advanced fraud detection or predicting disease outbreaks.
NLP, or natural language processing, sees AI trained on human language to understand and interact with it. Recent developments in how NLP models interact with text and speech data have led to the extraction of more valuable insights with a wide range of use cases, such as text summarisation, sentiment classification, grammar and spell checking, machine translation and natural entity recognition. There is significant overlap with generative AI here, given that generative AI is capable of many of the tasks NLP models also perform.
Transformer architecture, which processes language to understand the importance of each word and its relationship to others, means NLP models can measure context and relationships within text, leading to better summarisation and analysis as well as brand new applications. The transformer architecture, and specifically the success of large language models (LLMs) has meant that use cases which used to be siloed, e.g. text summarisation and sentiment classification, can now be performed in unison with a single machine learning system.
Advances in multimodal models have increased the scope of the field of Natural Language Processing to include data in the form of audio, speech, and images enabling Speech to Text, automated textual description of images, and more. Other NLP models focus on tackling bias. The data sets NLP models are trained on can often contain societal prejudices, so bias-aware NLPs seek to quantify and mitigate these biases. Evaluation frameworks are developed to assess the fairness and inclusivity of these models and promote fairness.
For businesses looking to process and make sense of vast amounts of data, NLP fast tracks manual tasks and leads users straight to the summarised output or analysis. Alleviating time has a transformative effect on how teams complete specific tasks, but the wider effect on the rest of their work shouldn’t be underestimated.
With advanced AI models producing insights based on relevant data, financial analysts can judge the sentiment of news data for stock market predictions, healthcare professionals can better digest medical literature for research and patient feedback, and chatbots across multiple industries can provide more accurate and valuable responses and process customer feedback. As NLP systems become more multimodal and sophisticated, the ability to interact with a system through text will become easier – for example, ChatGPT's 'plugins' feature - drafting and sending emails in a particular style, excel spreadsheets guided through text, etc
Significant advances in deep learning and neural networks have led to increased capabilities in object detection, image generation, video analysis and interpretation. Thanks to ever growing data sets of images from phones, security cameras and traffic systems, some computer vision models can now outperform humans in detecting, verifying and segmenting objects when analysing images.
Computer vision plays an integral role in the operation of autonomous vehicles and their ability to detect objects and take the necessary actions. There's also been a lot of attention and excitement about highly realistic and artistic production of synthetic images and videos, using Generative adversarial networks (GANs) and diffusion models. While part of Generative AI these also relate to Computer Vision.
Think of a computer vision model trained on a production system to identify defects in products and how it could outperform a human, and you can start to understand the power of computer vision.
Despite this, the general consensus is that there is a long way to go with computer vision models, but there is huge potential to improve tasks from detecting medical anomalies to virtually trying on products in online retail.
Computer vision holds the potential to instigate transformation across a number of regulated sectors:
Energy: Computer vision offers an expansive array of applications within the energy sector. A notable instance is in the realm of infrastructure inspection and maintenance, where drones, equipped with computer vision, can identify damage to power lines or renewable energy installations such as wind turbines or solar panels. This might also encompass predictive maintenance methods designed to detect anomalies in equipment and prevent failures.
By monitoring energy usage within buildings, it can aid in reducing unnecessary consumption by detecting and controlling energy wastage. It also plays a critical role in environmental monitoring, gauging the impacts of energy operations on wildlife and detecting oil or chemical leaks. Lastly, it contributes to safety monitoring within facilities, particularly in high-risk environments, ensuring the correct usage of safety equipment and restricting access to hazardous areas, thus aiding in accident prevention.
Healthcare: Computer vision is already making significant strides in the healthcare sector, with its techniques constantly improving and further refined, and playing an increasingly crucial role. It can assist in the analysis of medical images, enhancing the speed and accuracy of diagnoses, and providing support to medical personnel. It can also be applied to identify abnormalities in medical images such as X-rays, MRI scans, or CT scans. These techniques can assist in detecting diseases such as cancer, Alzheimer's, or heart conditions, often at earlier stages.
Transportation and Automotive: The use of computer vision in self-driving cars is already widely known. Autonomous vehicles are heavily reliant on computer vision to perceive their surroundings, detect obstacles, and navigate their environment. Furthermore, computer vision can also be used in areas such as traffic management, where it can analyse traffic flow, detect violations, or identify accidents.
In its first five days, ChatGPT acquired more than 1 million users. Investment into generative AI tools rose over 400% from 2020 to 2022. Bloomberg Intelligence predicts the market to be worth $1.3tn in the next 10 years. Safe to say, there is significant hype.
For the uninitiated, generative AI creates text, images and videos based on examples within its trained dataset. LLMs aren’t the only kind of generative AI – GANs and diffusion models have been popular for a while to create images, video and sound.
Wading through all the hype around generative AI can be difficult. We know generative AI has incredible potential to accelerate products and build solutions cheaper and more efficiently, while still maintaining quality. But organisations need to ask themselves how they can utilise it to make the business better. What is a specific challenge to increase revenue, reduce costs or save time? Where is the business lacking value?
The use of visual generative AI to produce synthetic images (for example through GANs and more recently diffusion models, which in quality is superseding the former) is worth a mention in a variety of fields, including regulated industries, where quality images might be difficult to come by. Synthetic datasets (e.g. in healthcare and the automotive industry) are much cheaper to produce. Generative AI such as GANs has also been used in drug discovery.
These same technologies are also creating new headaches for cybersecurity professionals, as they can be used to socially engineer attacks more easily, and therefore deep knowledge of the technologies and how they may be used is essential to counteract.
The benefits to improving operational efficiencies and accuracy are well documented but we see three key areas businesses can harness the power of generative AI:
For all the conversations we have with enterprise firms on their desire to adopt AI, ML and other sophisticated ways of using data, we invariably end up discussing their data foundations. No AI application will deliver true value to the business without reliable, accurate and democratised data. Often organisations want to begin their transformation journey at the end, not the beginning.
With a combination of modernised data platforms and modern techniques, organisations can create real value, be that for their customers, workforce or business – or sometimes all three. By embedding product thinking in the organisation and applying a scalable data governance approach to enable value generation through data, enterprise firms can build a solid basis which various AI and ML applications can feed off for long-term value to the business. A federated approach to data promotes accountability, transparency and accessibility, limiting bottlenecks in the subsequent production of AI products. The same approach can also cross organisational boundaries by enabling data marketplaces, where organisations can consume (and produce) high quality data across whole industries.
More often than not, what we need to focus on is getting the solid foundations in place to ensure businesses are confident in their assertions about adopting AI at scale. With a product thinking approach to data, creating these self-service infrastructures has profound implications for our customers ability to outperform competitors and achieve their desired outcomes.
Democratised access to deep actionable insights into your customers, your products and your business. These insights can then provide the best-possible ground for making killer business decisions and driving real competitive advantage.
At Mesh-AI, we're reimagining how enterprises harness data and AI for their own competitive advantage. Find out how we can help you leverage the data in your enterprise.
Find out how we leverage AI & ML to produce business systems that can make intelligent insights, choices and decisions.
Read our five point plan for building a rock solid AI strategy.