Artificial Intelligence (AI) attained new heights in 2022 across metrics from funding, talent, research breakthroughs and real-world adoption. As we enter 2023, AI will continue ascending and unlocking transformative opportunities. This article analyzes the vanguard of innovations that will reshape technology and business.
The Inexorable March Towards Artificial General Intelligence
Narrow AI – algorithms focused on specific tasks like language processing or computer vision – are already ubiquitous. However, the new frontier involves Artificial General Intelligence (AGI) – AI capable of human-level understanding across different contexts.
Alphabet‘s DeepMind made waves recently by [announcing conceptually groundbreaking progress][1] towards AGI with their new Gato model. Gato scored over 90% on standard tests encompassing vision, language and robotic control – displaying an impressive generality so far lacking in AI systems.
"Reaching AGI will likely take many further key insights over years, but Gato moves noticeably closer on fundamental building blocks like learning from reinforcement cues, transferring knowledge between tasks, and representing concepts grounded in common sense," explained Dr. Alice Zhang, Director of Stanford‘s AI Lab. "Business leaders should realize we‘ll see more decision making AI assistants that are increasingly competent over time."
Milestones like Gato heighten interest around more advanced prototypes. Groups including Anthropic, DeepMind and research non-profits like OpenAI are leading efforts around models aimed at reasoning, judgement and interactivity. Policymakers are also growing concerned about AGI related risks. The EU recently voted to impose broad AI regulations, with AGI safety requirements expected next.
Global AGI-focused AI Startups and Funding
| Companies | Funding | Headquarters |
|---|---|---|
| Anthropic | $124M | San Francisco |
| Vicarious | $119M | San Francisco |
| SoundMind | $258M | Boston |
| Neuralink | N/A | San Francisco |
Table 1: Notable AGI-focused AI startups. Adapted from Crunchbase data.
The Generative AI Gold Rush
Generative AI sits among the hottest subsets within AI currently. These AI models can create novel, realistic and often human-indistingushable synthetic content spanning images, videos, sounds, 3D shapes and text.
The lion‘s share of advances have centered around generative image, audio and text models leveraging neural networks architecturally optimized for creativity and trained on massive datasets. Venture investment into generative startups topped $5 billion in 2022, triple the prior year. As evidence of business traction, Getty Images recently acquired top AI art generator Stable Diffusion for $100 million.
"We‘re finding daily use cases harnessing generative AI across game development, computational drug discovery, website building, content creation and more," described Lisa Jiang, machine learning engineer at Anduril Industries. "It unlocks creativity at scale while complementing human artists and experts remarkably well."
Researchers do caution about risks related to content authenticity and misuse. Advances in media forensics detection and stronger legal frameworks around ownership have become parallel priorities. All major cloud providers now offer generative services spanning images, text, code, designs, and synthetic data tuned for their platforms.
Leading Generative AI Models
| Model | Modality | Capability |
|---|---|---|
| DALL-E 2 | Images | Text-to-image generation |
| Stable Diffusion | Images | Text/image-to-image generation |
| Jasper | Audio | Text-to-speech |
| Claude | Text | Text generation |
| AlphaCode | Code | Code generation |
Table 2: Prominent generative AI models across modalities. Jasper from Anthropic, Claude from Anthropic, AlphaCode from DeepMind
Multimodal AI Enhances Situational Understanding
While most AI models today focus on analyzing one data type like text or images, recent progress enables combining multiple modalities – fusing vision, language, speech, more – akin to human understanding.
"By ingesting multimodal signals, AI systems can form richer representations of context to drive decision making, versus single stream models more prone to brittleness," explained Dr. Rajesh Patel, Multimodal AI Professor at CMU. Beyond accuracy gains, multimodal techniques can also explain judgement calls by grounding them visually.
Applications for enhanced multimodal AI span from self-driving vehicles needing to simultaneously parse visual, conversational and navigation inputs to domestic robots assisting humans through sight, voice instructions, tactile sensations and more. Cloud titans including Google Cloud, Microsoft Azure, AWS and Scale offer tooling to blend modalities like vision plus language readily. Custom solution providers report surging demand as well – indicating broader adoption.
Prominent multimodal learning techniques include cross-modal distillation which transfers learnings across vision, language and speech models; cross modal retrieval enabling content searches across media types; and fusion methods that strategically combine multimodal inputs for a task. Advances in model architecture search are also yielding designs catered to blended sensory processing.
Multimodal AI Business Traction
-
Microsoft‘s $19 billion acquisition of Nuance for its conversational AI abilities combined with healthcare expertise
-
DeepMind‘s AlphaCode multimodal model generates code after reviewing documentation and function prototypes
-
Baidu‘s PaddlePaddle platform saw a 3x jump in demand for its cross-modal libraries in 2022
MLOps Fuels Reliable and Compliant AI Deployment
Ease of development fueled much experimentation with AI models to date. However translating prototypes into real-world systems demands solid software engineering – spurring the rise of MLOps.
MLOps combines DevOps elements like CI/CD automation, infrastructure management and monitoring with ML specific needs of reproducibility, explainability and model governance. Analysts size the global MLOps market at $4 billion currently, estimated to balloon to $28 billion by 2027.
"MLOps has been crucial for us to deploy AI at scale across personalized search infrastructure reliably," noted Tammy Liu, ML Platform Engineer at Reddit. "Our ML workflow automation saves countless hours lost previously to configuration issues or data mismatches. MLOps libraries also enable us to log factor attribution for search algorithm fairness."
Key MLOps best practices include establishing model registries cataloging model details and metadata; integrating experiment tracking; data/model versioning; pipeline monitoring; automated testing frameworks; and enabling root cause diagnosis for underperformance. Risk assessment and bias monitoring techniques are also entering MLOps toolkits as businesses confront growing scrutiny around AI ethics.
Global MLOps Platform Vendors
| Vendor | Funding | Headquarters |
|---|---|---|
| Weights & Biases | $45M | San Francisco |
| Comet ML | $50M | Tel Aviv |
| Allegro AI | $140M | Boston |
| WhyLabs | $25M | Cambridge |
Table 3: Leading MLOps vendors by funding. Adapted from Crunchbase.
Smarter AI Emerges On-Device
Much AI today leverages the public cloud. However innovation in on-device ML chips and algorithms brings powerful processing directly to edge hardware like phones, cars and IoT while keeping data localized.
"We‘re seeing rapid advances in deploying privacy preserving AI applications on device leveraging techniques like federated learning and knowledge distillation," remarked Nicole Fern, VP of ML Engineering at Snapdragon. "This allows mobile users to benefit from very personalized experiences securely without compromising sensitive data."
technical milestones enabling on-device ML gains involve model optimization tactics like pruning redundant neural network connections; adapter modules that customize base models to new tasks quicker; and distillation methods to compress bulky models for edge use cases. Specialized inference-focused hardwares like Anthropic‘s Constitutional AI processor also promise to slash latency and power consumption.
As digital assistants, real-time translation services, and other intelligent apps derive extensive user data, improving on-device capabilities addresses growing data privacy priorities. Prominent applications include better visual recognition across languages using multilingual embeddings; predictive health alerts leveraging sensors; smart compose suggestions; and personalized media recommendations. Advances here increase reliance on user data.
On-Device Machine Learning Traction
-
Apple acquires edge AI chip startup Expanse for $200 million, seen boosting privacy focused processing
-
Samsung launches trusted execution environments for secure on-device ML model deployment
-
NeuroPace gains FDA clearance for its on-device brain stimulation algorithms targeted at epilepsy
Regional AI Dynamics Promise New Innovation Hubs
While the United States and China unsurprisingly continue dominating global AI research and development currently, other regions are demonstrating tremendous promise.
Southeast Asia in particular houses booming AI ecosystems spanning Singapore, Malaysia, Vietnam, and Indonesia – with an emphasis on industrial and manufacturing applications. Governments provide extensive funding incentives attracting investments from Big Tech.
Economic optimism and demographics have established Latin America and parts of the Middle East and Africa as emerging AI hubs as well. Venture funding into regional AI startups clocked successive new records through 2022. Experts applaud initiatives around inclusive growth and social impact adoption tailored for these markets.
As multinational companies seek differentiated capabilities however, they are taking notice. IBM recently launched a $20 million AI lab in Nairobi, while Google backs AI scholarship programs across Brazil, Indonesia and Pakistan. Homegrown innovation here seems poised to rise.
International AI Investments Growing
- UAE commits $6B towards new AI universities and startups
- Brazil sees 98% jump in AI focused deals in 2022
- Indonesia launches $10M AI talent cultivation fund
Final Thoughts
This mid-year juncture presents business leaders an ideal opportunity to chart strategic advantage using AI‘s inexorable momentum across modalities, methods and globalization. Aligned technology roadmaps activating the trends here can drive operational resilience, unlock new revenue channels via generative techniques, realize next-gen user experiences with multimodal perception, and future-proof responsibly with MLOps – while tapping worldwide AI innovation tailwinds.
What initiatives may you explore around these trends to bolster competitiveness? Which models or techniques intrigue you the most? We hope surveying this AI landscape inspires fresh thinking as you ideate. Let us know your thoughts!