By Patrick Ott (ERNI Germany)
In the evolving landscape of AI, balancing the integration of advanced technologies with established methods is crucial. For software engineering companies, it is essential to be able to recognise the importance of leveraging the latest AI innovations like transformer models and valuing the efficiency of traditional AI techniques. This allows us to create solutions that are both forward-looking and grounded in proven methodologies, ensuring that we meet the needs of our partners in a way that drives meaningful progress while maintaining operational simplicity.
The release of ChatGPT marked a significant milestone in the AI megacycle that has captured the world by storm. The so-called attention mechanism laid the foundation for transformer models like GPT (Generative Pre-trained Transformer), which have revolutionised natural language processing (NLP) and extended to image and video processing. Today’s AI systems particularly excel in areas where humans face challenges, such as gaining a holistic understanding of large datasets and capturing the interdependencies within data items.
Real-world applications of AI models
AI is transforming industries by boosting operations and driving growth. For example, it predicts customer behaviour from vast datasets, enabling personalised marketing strategies and automating customer service through intelligent chatbots that improve response times. In supply chain management, AI forecasts disruptions and optimises logistics to maintain smooth operations.
Generative AI accelerates content creation, helping marketers develop ideas, conduct keyword research, and analyse competitors, all of which enhance campaign engagement. AI also optimises business processes, from HR to inventory control, driving efficiency and reducing costs. AI-powered chat agents and automation tools handle administrative tasks like scheduling appointments and handling emails, streamlining operations and improving job satisfaction.
Klarna, a global payments platform, leverages AI to dramatically enhance productivity while lowering costs. By using an AI assistant for customer service, Klarna performs the equivalent work of 700 full-time employees, handling two-thirds of customer queries and reducing resolution times by 80%. This automation has increased revenue per employee by 73% and saved the company $40 million annually. The integration of AI has helped the platform boost revenue by 27% while keeping operating expenses flat.
To mention a case from our own project pool, a partnership with a major transportation company aimed to optimise capacities through AI. By developing a machine learning algorithm, the project improved occupancy forecasts, helping to distribute passengers more efficiently across assets. Integrating diverse data sources into a cloud-based platform, and bringing expertise in machine learning and data engineering enabled better planning, load management and resource allocation, resulting in enhanced operational efficiency and customer experience.
Foundational models and their possibilities
Foundational models, large-scale AI systems trained on diverse data, are revolutionising the AI landscape. These models go beyond traditional language interpretation to include multi-modal capabilities, integrating text, images and other data types. For instance, OpenAI’s GPT-4 and Google’s BERT have demonstrated these multi-modal capabilities, opening new avenues for AI applications. Foundational models are being developed for various advanced applications, including autonomous driving, managing bureaucratic loads, and enhancing decision-making through querying other models.
The versatility of foundational models allows them to interact and complement each other, leading to more robust and informed decision-making.
Integrating these models across different domains significantly enhances their capabilities. For example, combining image recognition and language models enables detailed visual and contextual analysis. This cross-domain integration is crucial as robots perform more everyday tasks, promising autonomous performance in complex tasks across industries from manufacturing to healthcare.
One exciting application of foundational models lies in managing the increasing bureaucratic load within companies.
Imagine an AI that interacts seamlessly with both employees and customers, understanding and streamlining company operations while providing feedback for continuous improvement. This kind of foundational model will revolutionise business processes, increasing efficiency and productivity. The ability to create and deploy such versatile models will drive innovation and operational excellence across various sectors, opening new possibilities and improving outcomes in previously unimaginable ways.
The value of older AI methods
While transformers and generative AI are powerful, older AI methods still hold significant value. They excel in processing large datasets and handling tasks like automated warranty claims, risk assessment in banking, and predicting stock levels for logistics. Traditional models remain effective in the fraud detection, recommendation systems and predictive maintenance. Their development, training and operationalisation are often faster and cheaper than their transformer-based counterparts.
The software industry as well as the data science industry should leverage both new and old AI tools, applying the most efficient technology to each challenge. Existing business process automations should remain if they function well, aligning with the principle of Occam’s Razor: the simplest solution is often the best. Newer AI models can augment and enhance existing automation algorithms, increasing productivity without the need to start from scratch.
Standardised interfaces and data
One crucial aspect that companies should focus on is the standardisation of internal and external data interfaces, along with proper data description and cleansing.
Standardising data interfaces not only allows for scaling potential AI use cases but also accelerates the development of AI automation by reducing the manual data engineering required to make the AI work.
Imagine a standardised interface for booking data in the airline industry. AI can be employed for quick, text-based searches of bookings and for changing bookings using fully automated chat agents capable of interacting with customers in natural language, including speech. Once established and properly described, this data could be utilised to enable other use cases, such as analytics, or to scale the use case to partner airlines.
Considering the potential for entire industries to agree on specific standard interfaces – an initiative that could be championed by tech giants like OpenAI, Google or Microsoft – company-spanning microservices could become a reality. This would facilitate seamless integration and interoperability across various businesses, enhancing efficiency and innovation industry-wide.
Conclusion: Balancing innovation with proven methods
To effectively embrace the AI megacycle, balance cutting-edge technologies with proven traditional methods. By leveraging the strengths of both, industries can achieve greater efficiency, innovation, and security. Responsible AI implementation can transform industries and everyday life, maintaining trust and navigating the complexities of this new era.
For a deeper dive into how AI is transforming industries, explore our article on the fusion of AI and computer vision in modern manufacturing. Learn how these technologies drive efficiency and innovation.