Better Together: Humanity + Machine Learning

Speaker Context & Core Thesis

• Tanya Cushman (Reviewer) counters the dominant media narrative of mass unemployment ("robots coming for jobs") and existential AI risk ("Skynet") with a thesis that thoughtful machine learning deployment will make humans better decision-makers, extend lives, and strengthen society.
• The speaker notes a disconnect between front-page headlines focusing on AI threats and back-page/academic news focusing on practical integration of ML into software to augment human capabilities.

Global Adoption Trends & Investment

• The Neural Information Processing Systems (NIPS) conference sold out in 11 minutes and 38 seconds, indicating intense academic and researcher engagement.
• Paul Doherty (CTO of Accenture) states that AI is the fastest-growing technology trend in business history compared to the rise of client-server computing, the internet, and smartphones.
• Politicians globally are declaring their regions prime locations for AI, with billions of dollars in state spending directed toward building AI ecosystems.
• China has established a national, top-down goal to become the world leader in AI by 2030, supported by a detailed five-year milestone infographic distributed to citizens.

Category 1: Automating Routine to Leverage Human Strengths

• Amazon Logistics: By deploying over 100,000 Kiva robots to carry bins, Amazon hired 235,000 new employees in the last year; human roles shifted from manual carrying to creative problem-solving (e.g., unsticking robots).
• Customer Relationship Management (CRM): Systems capable of automatically populating CRMs from emails and calendars can optimize timing for outreach (e.g., calling prospects when they are most receptive), increasing the utility of sales data.
• Legal Discovery: The company Everlaw uses ML to sort and categorize massive document volumes; this allows lawyers to focus on the creative task of constructing legal narratives rather than routine evidence gathering.
• Call Center Monitoring: Dialpad uses real-time sentiment analysis on transcribed calls to advise agents on repairing negative interactions as they occur.
• Medical Documentation: Systems like Suki automate electronic health record entry during patient visits, allowing doctors to focus on bedside manner and empathy rather than typing, while improving diagnostic accuracy through algorithmic support.
• Automated Videography: Drone technology can automate camera positioning and framing for events, allowing human photographers to remain present in the moment rather than directing poses.

Category 2: "Superpowers" in the Physical World (Sensing)

• Voice Fraud Detection: PinDrop combats the 90 voice fraud attacks occurring per minute in the US by using voice prints and unique phone noise signatures to verify caller identity and location.
• Indoor Farming: Vertical farms using ML-driven sensors achieve yields of 15 pounds per square foot (compared to 1 in open fields) using 95% less water and no pesticides by adjusting environmental variables based on real-time plant growth data.
• Factory Training: Drishti analyzes video of factory workers to provide real-time guidance, resulting in a 25% increase in labor productivity and a 50% reduction in error rates for new hires in high-turnover environments.
• Construction Monitoring: Doxil uses robots to generate daily 3D scans of construction sites, comparing progress against blueprints to improve labor productivity by 38% and reduce costs (e.g., a project finished 11% under budget).
• Automotive Evidence: Tesla dashcam videos provide reliable, objective witnesses for traffic incidents (e.g., an intersection crash in North Carolina) to instantly resolve liability disputes where human testimony fails.
• Vision Assistance: OrCam devices clip onto glasses to provide text and face recognition, helping visually impaired individuals navigate daily life and remember names.

Category 3: Improving Human Decision-Making

• Bias Mitigation in Hiring: Textio uses a "gender meter" and regional analysis to optimize job descriptions, removing language that appeals disproportionately to specific demographics or locations.
• Sales Performance Coaching: Cresta.ai extracts best practices from top-performing sales reps to provide real-time guidance to lower-performing staff, increasing conversion rates by 20% and reducing training time by 66%.
• Responsible Mineral Sourcing: Cobalt Metals uses ML to analyze weather, plant, and geological data to identify cobalt reserves outside the Democratic Republic of Congo, where 40,000 children work in unsafe mining conditions.
• Alternative Credit Scoring: Branch Financial evaluates borrowers in regions without credit bureaus using mobile behavior patterns (e.g., battery usage, call timing, app presence) to predict loan repayment; notably, having gambling apps on a phone was correlated with higher repayment rates.
• Medical Diagnosis: In a Harvard/Cornell study, algorithms detected 92% of cancers, humans detected 96%, but human-algorithm teams detected nearly 100%, demonstrating that ML augments rather than replaces human diagnostic capability.

Category 4: Automating Dangerous Tasks

• Ocean Rescue: The "Little Ripper" drone performed the first ocean rescue by drone, locating surfers in distress and deploying flotation devices and shark repellent.
• Medical Logistics: Zipline delivers blood to 20–25% of Rwanda's transfusions via fixed-wing drones, replacing dangerous ground travel through difficult terrain.
• Military Operations: Shield.ai's "Nova" drone clears buildings, creating 3D maps and identifying friend vs. foe to protect soldiers from high-risk operations.
• Long-Haul Trucking: With 69% of long-haul truckers being obese and high rates of substance use due to monotonous, stressful work, the industry is moving toward automation to prevent physical health decline.
• Road Safety: Autonomous vehicle technology addresses the fact that road accidents are the leading cause of death for ages 15–29 and consume 3% of global GDP.
• Infrastructure: Companies like DeepMap, Intuition, and Syngin are developing 3D maps, simulation tools, and software to ensure safe deployment of self-driving vehicles.

Category 5: Enhancing Human Connection & Understanding

• Autism Support: Affective and BrainPower use video overlay to help children with autism recognize and modulate reactions to the emotional states of others.
• Real-Time Translation: Baidu demonstrated real-time speech-to-speech translation using predictive sentence completion to overcome the 20-minute attention limit of human interpreters, enabling seamless cross-language collaboration.
• Crisis Intervention: Analysis by Crisis Text Line and Stanford University found that the word "ibuprofen" is 14x more predictive of suicide than the word "suicide," and that counselors who immediately address the root cause are more successful than those focusing on rapport.

Strategic Recommendations for Implementation

• Data Integrity: Organizations must audit datasets to prevent the systematization of existing biases (e.g., gender discrimination); companies are creating new senior roles such as "Director of AI Ethics and Policy" to oversee fairness.
• Open Source Collaboration: The speaker argues against a nation-state AI arms race, advocating for open-source collaboration (citing Linux, Android, and TensorFlow) to accelerate progress and share benefits globally.
• Job Market Evolution: While routine jobs will be destroyed, the speaker predicts the creation of new jobs requiring "compassion" and "creativity," such as teachers, tour guides, and health coaches, which cannot be automated.
• Educational Reform: Curricula should be updated to prioritize kindergarten-style values of fairness, empathy, and "playing fair" to prepare a workforce capable of working alongside AI rather than competing with it.