I believe it is crucial to distinguish between genuine capabilities and overblown expectations. Here's a no-nonsense breakdown of AI's current abilities and limitations as of late 2024.

The "Hell Yeah, We Can Do That" List

Pattern Recognition and Prediction:

One of the things that current AI models are decently competent in is the identification of complex patterns in vast datasets. There are many examples of this from companies like Kensho Kensho and Numerai Numerai who utilize AI to analyse market trends and predict stock movements. In medical imaging, AI algorithms like Google's DeepMind have been trained to detect anomalies in scans with a high degree of accuracy. In a study published in Nature, DeepMind's AI system was able to detect breast cancer from mammograms with an 11.5% reduction in false positives and a 5.7% reduction in false negatives compared to human radiologists. This capability extends to so many other fields like retail analytics - Amazon has leveraged pattern recognition to analyse customer purchasing behaviour - and fraud detection where banks and financial institutions are implementing AI-driven pattern recognition algorithms to detect fraudulent transactions. It has even made its way into the world of sports with teams employing AI to analyse player performance and game strategies. Tools like STATS LLC STATS LLC utilize pattern recognition to evaluate player movements and team dynamics, providing insights that inform coaching decisions and game tactics. It can even provide value on a much larger scale with researchers at UC Berkeley developing an AI model that predicts heatwave patterns up to 50 days in advance by analysing historical temperature and pressure data.

Natural Language Processing (NLP):

NLP has transformed human-computer interaction and considering that what we currently call AI are just Large Language Models (LLMs), it is unsurprising that this is an area in which they are not only useful but exceptionally effective. On the well-known popular consciousness level we have virtual assistants like Apple's Siri Siri, Amazon's Alexa Alexa, and Google Assistant Google Assistant - all use NLP to understand and respond to voice commands, making it easier for users to access information and control devices. We also have Google Translate Google Translate which employs advanced neural machine translation (NMT) to facilitate real-time translation across 109 languages, extending to its camera feature that allows users to translate text from images instantly, which is particularly beneficial for travellers navigating foreign languages. Other notable services include Microsoft's Translator Microsoft's Translator and DeepL DeepL, with the latter often recognized for its superior handling of idiomatic expressions. The obvious extension of these capabilities is into the world of customer service chatbots, which are increasingly used in customer service to handle inquiries and provide personalized recommendations. A case study by Autodesk Autodesk highlighted an 85% reduction in resolution time for customer inquiries after implementing a chatbot, alongside a 10% increase in customer satisfaction. Other examples of where current AI models excel in NLP-focused tasks are in the fields of sentiment analysis and text summarization. In the former companies like Brandwatch Brandwatch and Hootsuite Hootsuite leverage NLP for sentiment analysis, which helps businesses monitor social media posts and customer reviews to gauge public opinion. In the latter, there are startups like Primer Primer and Scribbr Scribbr that use NLP for text summarization, allowing users to quickly digest lengthy documents such as news articles or research papers. In the legal sector, tools like Casetext Casetext and Legal Data Intelligence Legal Data Intelligence use NLP to analyze legal documents, assisting lawyers in research by generating concise summaries.

Personalization and Recommendation Systems:

The last of what I believe are current AI's core competencies is the ability to not only analyse user behaviour but also use this data to provide hyper-personalized experiences. We have all experienced this through streaming services like Netflix and Spotify’s content recommendations Netflix and Spotify’s content recommendations as well as product recommendations from e-commerce brands like Amazon and Alibaba. These recommendation systems produce results too; Netflix reports that 80% of watched content comes from its recommendations. A McKinsey study found that 35% of Amazon's revenue is generated by its recommendation engine, and Spotify's Discover Weekly playlist delivers more than 5 billion tracks to more than 40 million users through their personalised playlists alone. Personalisation and recommendation have been taken even further with products like Flipboard, which personalises news feeds using machine learning algorithms that analyse user reading habits, and E-learning platforms like Coursera use AI to personalize course recommendations based on users' interests, past courses, and learning pace. Travel websites such as Expedia utilize AI-driven personalization to recommend destinations, hotels, and activities based on previous searches and bookings. Fitness applications like MyFitnessPal MyFitnessPal use AI to personalize diet plans and workout routines based on user input regarding fitness goals, dietary restrictions, and activity levels.

The "Nice Try, But No Cigar" List

True General Intelligence:

Despite impressive advancements, Artificial General Intelligence (AGI) remains elusive. AGI refers to a machine's ability to understand, learn, and apply knowledge across various topics, much like a human. While AI systems like DeepMind's AlphaGo have mastered complex games like Go and chess, they are limited to narrow, very specific tasks. They lack the adaptable reasoning and common-sense knowledge characteristic of humans. They cannot transfer their learning from one topic to another or handle tasks outside their trained scope. For example, an AI system trained to recognize images cannot suddenly start understanding natural language or solving mathematical equations without extensive retraining. It is also very important to highlight that while language models like GPT can generate human-like responses, they do not possess genuine understanding or consciousness. They are essentially pattern-matching machines that predict the most likely next word based on vast amounts of training data. They do not comprehend the meaning behind the words they generate or have any real-world understanding. This lack of genuine intelligence limits the ability of current AI systems to handle the complexity and unpredictability of real-world situations, which often require a combination of knowledge, reasoning, and adaptability.

Ethical Decision-Making:

We should never put a currently existing AI model in charge of a task that requires ethical decision-making. Current AI models operate on algorithms and data and lack inherent moral judgment. While they can be trained to make decisions based on predefined rules and guidelines, they cannot autonomously navigate complex ethical dilemmas or consider broader societal implications. For example, an AI system designed to optimise resource allocation in a hospital may prioritize efficiency over fairness, leading to unintended consequences like discrimination against certain patient groups. The challenge of ethical decision-making by AI is compounded by the fact that ethics are often context-dependent and can vary across cultures and situations. What may be considered ethical in one scenario may not be appropriate in another. Moreover, AI systems can and do perpetuate and amplify human biases present in their training data, leading to unfair or discriminatory outcomes. The responsibility for ethical oversight remains firmly in human hands, requiring ongoing monitoring, adjustment, and intervention to ensure that AI systems align with human values and societal norms.

Creative Insight and Innovation:

Current AI models can generate content based on existing patterns, and this does produce some solid results, but true creativity - involving novel, abstract thinking - remains beyond its current capabilities. While AI systems like GPT-3 can produce human-like text, art, and music, they are essentially recombining and adapting patterns from their training data. They lack the ability to generate truly original ideas or think outside the box in the way that humans can - ie they cannot create new patterns, only mix match and remix ones they have current knowledge of. The truth is that groundbreaking innovations and paradigm shifts still require human ingenuity. While AI can assist in the creative process by generating starting ideas or optimizing designs, it is not capable of the kind of abstract, imaginative thinking that drives true innovation.

As the AI landscape is rapidly evolving, presenting both exciting opportunities and important limitations, we at Cerebral Circuit are committed to helping businesses navigate this complex terrain. Understanding AI's strengths and limitations is crucial for setting realistic expectations and effectively leveraging its power. As we continue to push the boundaries of what's possible, it's essential to approach AI integration strategically, balancing technological capabilities with human insight and creativity - this list will be updated!