Discover more from Curious Compass
Artificial Intelligence in Business
How AI is shaping Profits and Processes
The AI market is expected to grow from $81 billion in 2022 to $909 billion by 2030, with a 35% CAGR from 2022–2030.
Of the top five cutting-edge AI technologies currently being spotlighted, generative AI is growing the fastest. You can find a comprehensive briefing on generative AI in the first edition of AI in Business.
Generative AI is anticipated to become a threat to all businesses in various sectors in the forthcoming years. The influence will broaden across industries and business operations as this type of AI improves in accuracy, offering dependable factual guidance.
Below, you will discover a variety of use cases from diverse sectors, demonstrating the most innovative AI technologies. These include customer assistance chatbots, the development of news and marketing content, coding, and product design, and it's likely these uses will multiply in the future.
Advanced AI capabilities
When building AI systems, a paramount question that arises is: What capabilities or behaviors make a system intelligent?
Curious Compass is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
AI can be defined as any machine-oriented system having the capacity to perceive its environment, pursue objectives, adapt to feedback or alterations, provide information, take action, or even exhibit self-awareness and consciousness.
Evidently, beings like humans and animals can engage with their surroundings, adjust to changes, and fulfill their goals and survival. The notion of animals possessing ethics or moral reasoning is still under discussion; however, they certainly showcase sentience.
These AI capacities are ordered into five, namely:
Creation, also referred to as 'generative AI', encapsulates the capabilities of creating various types of content, like audio, video, and text, to things like software code, molecular structures, and even other AI (self-creation).
An interesting observation, Whereas AI researchers might be fascinated with creating AI systems with sentience and ethical awareness, there's a stronger business case for robots that just possess effective interaction, moderate decision-making abilities, and motion capabilities.
The latest trends in Generative AI
Most AI models are trained by a method known as reinforcement learning from human feedback (RLHF), where human moderators rate the AI’s output and provide feedback. This method is both time-consuming and requires a lot of effort.
As colorfully illustrated by systems such as ChatGPT and GPT-4, AI, particularly text-generating AI, has massive flaws. Because it’s often trained on questionable internet sources (e.g. social media), it’s often biased in obviously sexist and racist ways. And it hallucinates — or makes up — answers to questions beyond the scope of its knowledge.1
Constitutional AI describes a method of training an AI system according to a set of rules (or constitutions) predetermined by human operators. The AI system can evaluate its output against its pre-defined constitution and refine its output accordingly.
For example, Anthropic’s model “Claude” has been fine-tuned with constitutional training with the goal of becoming helpful, honest, and harmless. You can learn more about constitutional training here.
The Future of Work
The surge in public interest in AI has reignited concerns about job automation and the replacement of traditional professions.
Developments in generative AI mean that jobs that were once thought to be immune, such as content writers, software developers, and graphic designers, now also face the risk of replacement.
Many have called for further regulation of AI to limit its impact on jobs, with AI implementation being a central issue in the recent labor disputes in the film and TV industry.
Scientists at UC Berkeley developed a pioneering technique called Dynalang. They sought to expand AI models to understand their environmental context. According to its creators, Dynalang is capable of modeling the world through visual and textual inputs, similar to the multi-modal and self-supervised learning of humans. Future developments of this technique can promote society toward advanced models interacting with humans in different contexts.
A significant hurdle in AI research is empowering AI models to engage in natural conversation with humans. Current AI solutions, like Google's PaLM-SayCan, can comprehend straightforward directives like "get the blue block". However, they come short when faced with more intricate language contexts, such as knowledge transfer ("the top left button shuts down the TV"), situational updates ("we're short of milk"), or cooperative tasks ("the living room has already been cleaned").
For instance, if an AI agent hears "I've put the bowls away," its reaction should differ based on the task at hand: If it is in the middle of washing dishes, it should proceed to the subsequent sanitizing step; if it's serving dinner, it should get the bowls.2
AI Trust, Risk, and Security Management (AI TRiSM) represents a contemporary AI trend that advocates for data governance throughout corporations. It's primarily targeted at guaranteeing that organizations adhere to pertinent data privacy laws and guidelines.
By using AI TRiSM, organizations can improve their essential decision-making processes along with the security, interpretability, efficiency, and privacy of their processes driven by data.
AI development is dominated by US and Chinese companies
2023 YTD has seen a 5% increase in AI deals compared to all of 2022
Debt financing remains the most popular way for AI companies to raise capital
AI Use cases by sector
AI in Media
Adobe has introduced the Generative Fill feature in Photoshop, leveraging Firefly's generative AI. This feature enables users to adjust images using simple text prompts while maintaining the original image's quality.
US-based creative AI company, Seyhan Lee, has launched Cuebric, an AI tool poised to revolutionize film production. Cuebric streamlines complex processes for filmmakers, enhancing the user experience. Developed with XR Studios, this tool will first cater to specific groups, including LED volume stages and concept artists.
Hearst Newspapers, under Hearst Communications, has adopted generative AI tools, specifically ChatGPT, to enhance their journalism techniques. Such AI integrations help media outlets improve various facets of content production and distribution.
GetGenius, an American tech startup, has unveiled getgenius.ai, an AI-centric platform for social media management. This tool provides an all-inclusive solution for businesses and creators, optimizing digital marketing strategies and expanding their digital footprint.
AI in Consumer goods
AI in Product: Rembrand
Rembrand, backed by cosmetics giant L'Oréal, uses generative AI to integrate product visuals into videos for enhanced advertising.
Operational Optimization at Unilever
Unilever harnesses AI tools, including neural networks and the GPT API, to streamline business operations, from email categorization to food waste reduction, driven by its commitment to evolving consumer preferences.
Virtual Fashion by Photo AI
Swedish startup Photo AI has developed a tool that allows users to virtually try clothes on a digital self, merging AI with fashion insights to revolutionize the shopping experience.
Coca-Cola's AI Venture: Create Real Magic
Coca-Cola's Create Real Magic, created with OpenAI and Bain & Company, utilizes AI for inventive advertising, drawing from its archives, and is notably the first to incorporate OpenAI’s GPT-4 and DALL-E.
AI in Industrial
Toyota's AI Approach
Toyota Research Institute uses generative AI to enhance vehicle design. Starting with open-source AI tools, they now streamline design and engineering in vehicle creation.
Retrocausal's AI for Manufacturing
US startup Retrocausal has launched LeanGPT for manufacturing. They also offer Kaizen Copilot to help engineers refine assembly processes.
Autodesk's Forma optimizes workflows for infrastructure projects. Combining automation and AI, it assists architects in design and construction site analysis.
Hitachi Energy's Solution
Swiss-based Hitachi Energy presents the AI Hitachi Vegetation Manager to monitor vegetation near power lines, aiming to prevent wildfires.
AI in Transport
Cerence's Enhanced In-Car Assistant
Cerence, an automotive AI startup, has upgraded its Car Knowledge platform, using AI to improve interactions between drivers and in-car systems.
Nvidia's Supply Chain Solution: ReOpt
Nvidia's ReOpt is an AI service to optimize supply chain routes, addressing disruptions and targeting sectors like transportation and retail.
Argo AI's Driverless Trials
Argo AI from Pennsylvania is testing autonomous driving in Miami and Austin using Ford Hybrids, pushing for Level 4 autonomous tech in the US.
Avikus' Autonomous Nautical Milestone
Avikus, with SK Shipping, completed a transoceanic journey using the Prism Courage LNG vessel with Level 2 autonomous navigation, marking a global first.
AI in Healthcare
‘Dr GenAI’: Medical Assistance Chatbot
The innovative medical chatbot, 'Dr GenAI', uses OpenAI’s ChatGPT to provide quick, personalized medical advice based on individual health profiles. Developed with the AI capabilities of the CERVAI platform, it can analyze vitals, lab results, and physical attributes.
Paige’s Enhanced Diagnostic Suite
New York-based startup, Paige, has launched the Paige Breast Suite, a set of AI-driven tools aimed at aiding pathologists in the accurate diagnosis of breast cancer, reducing manual inconsistencies, and enhancing diagnostic accuracy.
Ganymed Robotics: Robotic Innovation in Orthopedics
Parisian startup, Ganymed Robotics, offers a state-of-the-art robotic aid for orthopedic surgeries. Combining computer vision and mechatronics, its flagship application is a surgical robot for total knee arthroplasty.
Sanofi's AI-Infused Application
French pharmaceutical giant, Sanofi, has introduced an AI-driven app aimed at transforming its operations from manufacturing to R&D, emphasizing streamlined processes and improved decision-making.
AI in Finance
Morgan Stanley's AI Tool
Morgan Stanley is utilizing ChatGPT-4, in collaboration with OpenAI, to efficiently retrieve analyst insights for its wealth management staff.
Salesforce's Financial Platform
Salesforce has launched a platform integrating generative AI and real-time data, enhancing financial customer interactions through its cloud services.
Modern Life's Insurance Chat Tool
Modern Life's AI chat tool aids insurance advisors, having demonstrated its expertise by passing a certification exam, streamlining client consultations.
Lightning Labs' Bitcoin Toolkit
Lightning Labs has released tools, integrated with OpenAI's GPT series, allowing AI applications to streamline Bitcoin transactions.
Environmental Impact of AI
AI's energy demands can impact the environment, but it also offers solutions to climate issues. Modern AI hardware and data centers have a significant carbon footprint, with data centers consuming nearly 1% of global electricity. As AI investments grow, so will energy use. However, there are initiatives to power AI with renewables and design efficient cooling systems. Additionally, AI aids in optimizing supply chains and building operations to minimize emissions.
LLM Carbon Emissions
The AI Index Report compared four LLMs, calculating their individual carbon emissions during training. OpenAI’s GPT-3 model emitted the most carbon out of all the models assessed. It was even higher than Gopher, an open-source model trained on 280B parameters. Multilingual language open model Bloom, with similar parameters to GPT-3, produced 25 tons of carbon in 2022, which was 20 times lower than GPT-3. Meta’s open, pre-trained language model OPT consumed the least power, with 1/7th the carbon emissions produced by GPT-3. It is also worth noting that this study did not include GPT-4, which could be much higher in terms of emissions.
Emerging Generative AI delivery models
The landscape of generative AI models for enterprise customers is evolving very rapidly, with multiple combinations of business models and technology architectures. As such, it is often difficult to compare offerings or establish their suitability for given use cases.
There are at least 5 factors that are critical, and at least 3 delivery models emerging.
The large language model itself: is it proprietary or open source? Is it licensed, rented on a pay-as-you-go basis, or simply developed in-house?
Pre-training execution: who runs and pays for the pre-training of a model like ChatGPT or LLaMa? The supplier or the enterprise customer?
Trained image data: is it proprietary and undisclosed, open source, or licensed? This is important as customers in industries with highly sensitive and valuable information would prefer to take a pre-trained model and run it internally on a private cloud rather than using it through an API in a public cloud setting.
Additional domain-specific training: does your use case require significant additional training on domain-specific data and/or knowledge? Using an out-of-the-box ChatGPT to implement a customer service bot probably does not require significant additional training, but using it to implement an insurance robo-advisor would require significant additional training on insurance knowledge as well as the customer-specific product offerings.
Proprietary data access security: how important is it to a specific enterprise customer that the data remain highly secure and private? API-based delivery models such as OpenAI's may not be suitable for, say, a pharma company wishing to use the LLM for writing highly confidential clinical trial protocols for a new drug.
Based on the most logical combinations of these factors, I believe generative AI solutions are likely to evolve towards similar economic and delivery models of cloud computing, split into 3 categories(hosted or SaaS, hybrid or licensed pre-training, and Organic or in-house) .
Cost comparison of Generative AI delivery Models
Organizational costs for AI models encompass financial expenses, team-building for AI expertise, and operational costs. Generative AI solutions are trending towards cloud computing economic models. When choosing between in-house infrastructure and public cloud services, factors like transactional volume and workload predictability are considered. Public clouds are favored for unpredictable workloads due to their pay-as-you-go pricing. Yet, companies like Dropbox and Netflix have moved predictable workloads in-house.
For generative AI, hosted models like OpenAI’s ChatGPT are popular, but due to high usage costs, companies with significant IT budgets may opt for hybrid or licensed models. A few may even develop in-house models, like Bloomberg's BloombergGPT, for minimal usage costs. Additionally, data perspectives are also crucial.
From a data viewpoint, the choice of AI delivery model depends on:
1. The need for domain-specific knowledge
2. Data sensitivity
Hosted or SaaS models like OpenAI are ideal. However, if there's sensitive data, a hybrid or licensed pre-training model is better since it ensures data safety within a private cloud. For intricate organizational and process knowledge, in-house deployment might be the best fit, but it's feasible only for well-resourced enterprises.
Recent advancements in machine learning, thanks to enhanced algorithms and greater computational capacity, have enabled AI to address real-world challenges.
By 2030, the entire AI market will surpass $900 billion.
GenAI is expected to grow at least 80% CAGR.
In the initial stages, AI investments will primarily focus on computer vision and chat platforms.
Early adopting verticals will be IT, retail banking, government, manufacturing, retail, and healthcare.
AI Tool Report
Learn AI in 5 Minutes a Day We'll teach you how to save time and earn more with AI. Join 70,000+ free daily readers for trending tools, productivity-boosting prompts, the latest news, and more.