Understanding AI models

Different types of AI models  

Machine learning models  
Machine learning (ML) is a subset of AI. While all ML models are AI, not all AI models use ML. The goal of AI is to enable machines to act in humanlike ways, whereas ML focuses on teaching machines to make decisions and predictions without explicit programming. ML models identify patterns in data, enabling them to learn and improve performance over time. 
  
Deep learning AI models  
Deep learning models, also known as deep neural networks, are advanced forms of ML. Inspired by the structure and function of the human brain, these models process large amounts of unstructured data, such as text, images and sounds. They excel at recognizing patterns to generate insights and predictions. Common applications include facial recognition, natural language processing (NLP), virtual reality and autonomous vehicles.

Generative AI models  
Generative AI (GenAI) refers to AI that can create new content, such as text, images, music, videos, translations and code. These models are trained on vast datasets and use deep learning to identify and analyze patterns, enabling them to generate original outputs. Examples include ChatGPT for conversational AI and DALL-E for text-to-image generation. GenAI has revolutionized the AI landscape, expanding its applications for businesses and the public.

Language models  
Language models are AI systems designed to understand and generate human language. The most advanced type, large language models (LLMs), is a subset of GenAI. LLMs are trained on massive natural language datasets using advanced ML techniques. They can generate nuanced and contextually relevant text responses to prompts. Examples include ChatGPT, Claude, Gemini, Microsoft Copilot and Meta AI.
 
Predictive AI models 
Predictive AI models leverage AI and ML to identify patterns, predict outcomes and generate forecasts through statistical data analysis. While predictive analytics isn’t new, AI enhances the speed and accuracy of these processes by utilizing large datasets. Applications include inventory management, customer behavior analysis, risk management and forecasting future trends.
 
Computer vision AI models  
Computer vision AI models use ML to train computers to interpret and understand visual data, akin to human perception. These models analyze images and videos to identify patterns and classify objects. Applications range from facial recognition and autonomous vehicle navigation to medical imaging. While DALL-E and DALL-E 2 involve image generation, they primarily combine computer vision and NLP.
 
Recommender AI models 
Recommender AI models analyze user behavior using big data analytics and ML algorithms to suggest items of interest. Commonly used in platforms like Netflix, Spotify and social media, these models personalize user experiences based on data such as past purchases, search history and demographics.

How AI models work

The process of creating an AI model that can understand, interpret and derive insights from data involves several steps: 

• Data gathering: Data forms the foundation for all AI projects, so collecting data is crucial — whether it’s a simple piece of text or a complex dataset.  
• Data cleaning and preparation: Collected data must be cleaned and prepared before an AI model can use it. Unnecessary, irrelevant or false data is excluded, and the data is formatted so the AI can use it. Data quality is critical, directly affecting a model’s accuracy and reliability. 
• Training: The AI model is trained by feeding the data into a selected algorithm, enabling the model to learn and refine for improved performance. 
• Testing: After training, the model is evaluated for accurate responses. 
• Fine-tuning: If a model isn’t delivering the desired quality of outputs, it needs to be adjusted and fine-tuned until it meets expected standards. 
• Deployment: Once a model is ready, it can be implemented to use within the organization. 
• Continuous improvement: As they work with new data, AI models can learn and adapt, allowing them to improve performance over time.

AI business use cases

Businesses use AI for a wide range of applications, with more being developed all the time. Common use cases include: 
 
Customer service: This is one of the most well-known AI applications. Many companies are already using chatbots and virtual assistants to help customers with questions and troubleshooting. These are becoming more sophisticated, allowing them to take on more tasks, freeing up human agents to tackle more difficult situations. AI can also be used for personalization and recommendations, helping customers find new products or services based on their history.  
 
Productivity and efficiency: AI is very good at automating tedious and time-consuming tasks to save time and resources and enable humans to focus on higher-value work. AI can also analyze current processes to identify performance gaps or bottlenecks, suggest ways to improve workflows and use data-driven prioritization to boost efficiency. 
 
Supply chain management: AI can be used to automate and improve many processes within supply chains for greater efficiency and better customer service. For example, AI applications help make demand forecasting more accurate, optimize inventory, monitor production, automate shipping and reduce downtime with predictive maintenance and troubleshooting. 
 
Content creation: GenAI tools empower people to create high-quality written, visual and musical content with natural language prompts. Examples range from writing, editing and proofreading to graphic design, image and video creation and editing and interactive storytelling. AI can also assist in creating code and debugging. 
 
Risk mitigation and security: AI can be used to reduce risk and protect important physical and virtual assets. AI is already key for cybersecurity measures including identifying system vulnerabilities, monitoring operations and stopping threats. AI can also be used to detect potential fraud and manage data for regulatory compliance. 
 
Innovation: The automation and efficiency-driving aspects of AI help streamline processes and allow people more opportunity to ideate and strategize. AI can also help accelerate R&D, design new products and optimize marketing and sales efforts. And AI-driven data analysis can help businesses see new opportunities and stay competitive. 
 
These are only a few examples of the ways in which AI is being used in business. New applications are being launched all the time — and businesses are discovering how to customize AI for their unique needs. 

Ethical and societal implications

While AI has great potential to help humans, it also has potential to harm, and people developing AI have a responsibility to prevent harmful outcomes. Areas of ethical and societal concern related to AI models include: 

• Accuracy: GenAI models are known to generate “hallucinations” — outputs that are false or simply created out of nowhere rather than based on factual data.  
• Bias: If models are trained on datasets that contain bias, that bias can become encoded in the model. In this way, societal biases such as racism can be perpetuated in AI outputs. 
• Digitally forged content: Malicious actors can use AI models to create deepfakes that can cause personal harm or be used in cybercrimes.  
• Copyright: AI models often use publicly available content without consent of its owners, and copyright and plagiarism issues arise because it’s difficult to trace how an AI model uses content. 
• Privacy: Sensitive personal data used to train AI models needs to be protected. 
• Transparency and accountability: The decision-making process of AI models is often unclear, making it difficult to evaluate outputs and assign accountability.  

AI ethics standards, such as UNESCO’s Recommendation on the Ethics of Artificial Intelligence, offer guidance for organizations seeking to develop and manage AI models in an ethical, responsible way. A growing number of governments are also developing legislation to regulate AI. 

Future trends in AI modeling

AI’s journey is just starting, and AI modeling will continue to evolve. A few trends that organizations should keep on their radar include: 
 
Agentic AI: Agentic AI consists of “agents” that can perform tasks for another entity autonomously. While traditional AI systems rely on inputs and programming, agentic AI models are designed to act more like a human employee, understanding context and instructions, setting goals and independently acting to achieve those goals while adapting as necessary, with minimal human intervention. These models can learn from user behavior and other sources beyond the system’s initial training data.
 
Multimodal AI: Multimodal AI refers to systems that process and generate content across multiple data modalities, such as text, images and videos. While many current AI models specialize in a single modality, advancements are enabling systems to integrate and transition seamlessly between them. For example, certain AI models can generate images from textual descriptions or create videos based on textual or visual inputs. These capabilities enhance user interaction by offering greater flexibility and intuitive applications. 
 
Closed source AI: The most well-known AI models, such as ChatGPT, DALL-E, Claude, Google’s Gemini and Microsoft Copilot, are proprietary models. These closed third-party vendor models are trained on huge amounts of data — something few companies have the resources to accomplish — and they are very powerful. However, they do have drawbacks. Governance can be a concern due to their “black box” style, which makes it difficult to oversee how they create outputs. Companies may also be justifiably wary about releasing sensitive data and IP into a system owned by another company.  
  
Open source AI: Open source models provide an alternative that requires far less resources than creating and training an LLM. Open source models are often free of charge, allowing companies an opportunity to build on existing code. These models allow more oversight and customization via fine-tuning than proprietary models and can be available to an organization indefinitely. Companies can use open source models to keep their data private while using the power of AI to create value from that data in customized applications. Open source models democratize AI — and use of these models will continue to grow.  

For example, Databricks offers DBRX, a general-purpose LLM that enables customizable, transparent GenAI for companies of all sizes. DBRX serves as a starting point to be fine-tuned or adapted for specific AI applications. DBRX outperforms all established open source models on standard benchmarks.

Data-centric AI with Databricks

The ability to manage AI models has become critical for enterprises to stay competitive. Mosaic AI, part of the Databricks Data Intelligence Platform, unifies data, model training and production environments in a single solution. This allows organizations to securely use enterprise data to augment, fine-tune or build their own ML and generative AI models. With Mosaic AI, organizations can securely and cost-effectively build production-quality AI systems, centrally deploy and govern all AI models and monitor data, features and AI models in one place.