00:00:00;29	00:00:01;16	>> Dr. Alex Kirkpatrick: Hello.
00:00:01;18	00:00:03;13	I'm Dr. Alex Kirkpatrick
		from the Center
00:00:03;15	00:00:05;27	for Sustaining Agriculture
		and Natural Resources
00:00:05;29	00:00:08;19	at Washington State University.
00:00:08;21	00:00:12;01	If AI is given the agency and
		autonomy to do what humans used
00:00:12;03	00:00:14;20	to do in agriculture,
		how do we then attribute
00:00:14;22	00:00:16;23	moral responsibility?
00:00:16;25	00:00:20;05	Can a machine be held
		responsible for its actions?
00:00:20;07	00:00:24;04	If a machine learns from
		data provided by a society
00:00:24;06	00:00:27;08	with a long history of
		bias and discrimination,
00:00:27;10	00:00:31;09	does that machine give biased
		and discriminatory outputs?
00:00:31;11	00:00:35;26	This video explores the nascent
		and evolving field of AI ethics
00:00:35;28	00:00:38;06	and the philosophical
		questions that we,
00:00:38;08	00:00:40;05	as a culture, may
		have to answer.
00:00:40;07	00:00:43;13	Let's explore.
00:00:43;15	00:00:59;15	[ Music ]
00:01:03;15	00:01:06;21	Ethics matter because they
		protect against exploitation,
00:01:06;23	00:01:10;02	support fairness, and
		build long-term trust,
00:01:10;04	00:01:13;00	all crucial if AI is
		going to be accepted
00:01:13;02	00:01:14;29	and effective in agriculture.
00:01:15;01	00:01:18;16	As society grapples with the
		ethics of AI and affording power
00:01:18;18	00:01:21;24	to autonomous machines, so
		must agriculture contend
00:01:21;26	00:01:25;23	with large philosophical
		and moral quandaries.
00:01:25;25	00:01:28;13	People are often
		concerned about their data
00:01:28;15	00:01:30;26	above all else when
		it comes to AI.
00:01:30;28	00:01:33;04	Who owns the data
		collected from farms
00:01:33;06	00:01:35;11	and agricultural operations?
00:01:35;13	00:01:38;09	This is a big question.
00:01:38;11	00:01:41;08	AI often exists in a
		so-called black box,
00:01:41;10	00:01:45;00	meaning it's not entirely clear
		how it comes to decisions.
00:01:45;02	00:01:48;15	Even if we could always
		trace input to output,
00:01:48;17	00:01:51;29	does the average AI user in
		agriculture have the know-how
00:01:52;01	00:01:54;08	to interrogate the algorithm?
00:01:54;10	00:01:59;17	Should the user have the ability
		to override an algorithm?
00:01:59;19	00:02:03;01	Who reaps the rewards
		of AI in agriculture,
00:02:03;03	00:02:06;21	and who experiences
		the brunt of the risks?
00:02:06;23	00:02:09;03	How are these risks
		and rewards distributed
00:02:09;05	00:02:13;22	across agricultural operations
		and surrounding communities?
00:02:13;24	00:02:17;28	Relatedly, there exists a lot of
		socioeconomic inequity already
00:02:18;00	00:02:22;00	in society broadly, which
		extends to agriculture.
00:02:22;02	00:02:26;16	Does AI widen or shrink existing
		gaps between the diverse groups
00:02:26;18	00:02:30;29	of people that make up
		agricultural communities?
00:02:31;01	00:02:34;24	Training and maintaining AI
		is very resource-intensive,
00:02:34;26	00:02:37;18	demanding a lot of energy
		to power data centers,
00:02:37;20	00:02:40;07	supercomputers, cooling
		systems, hardware,
00:02:40;09	00:02:42;02	manufacturing, and more.
00:02:42;04	00:02:43;13	How do these impacts weigh
00:02:43;15	00:02:46;29	against the potentially positive
		impacts on precision agriculture
00:02:47;01	00:02:48;10	and resource management?
00:02:48;12	00:02:52;17	Furthermore, AI might
		automate tasks that humans are,
00:02:52;19	00:02:54;24	at the moment, paid to perform.
00:02:54;26	00:02:57;23	What impact does this have
		on labor and the lives
00:02:57;25	00:03:00;21	of the individuals affected?
00:03:00;23	00:03:04;04	There are doubtless many more
		questions like these that are as
00:03:04;06	00:03:08;23	yet unresolved and demand
		attention and deliberation.
00:03:08;25	00:03:11;12	Given the pace of
		technological development,
00:03:11;14	00:03:14;14	the urgency of answering
		such questions has led
00:03:14;16	00:03:19;21	to many AI ethics
		initiatives and policies.
00:03:19;23	00:03:22;06	The Organization for
		Economic Cooperation
00:03:22;08	00:03:25;03	and Development is an
		intergovernmental organization
00:03:25;05	00:03:26;24	with 38 member countries,
00:03:26;26	00:03:29;28	including the United
		States and most of Europe.
00:03:30;00	00:03:33;12	Its aim is to stimulate economic
		progress and world trade.
00:03:33;14	00:03:36;28	And they were the first
		intergovernmental organization
00:03:37;00	00:03:41;13	to adopt standards in
		AI ethics back in 2019.
00:03:41;15	00:03:42;22	Their first principle is
00:03:42;24	00:03:46;05	that ethical AI should promote
		prosperity for both people
00:03:46;07	00:03:49;08	and the planet, enhance
		inclusion,
00:03:49;10	00:03:53;24	environmental sustainability,
		and reduce inequalities.
00:03:53;26	00:03:55;15	In agriculture, this
		could translate
00:03:55;17	00:03:59;00	to ensuring small-scale and
		indigenous farms benefit as much
00:03:59;02	00:04:02;08	from AI diffusion as
		larger operations,
00:04:02;10	00:04:05;05	or that the positive
		environmental impacts
00:04:05;07	00:04:08;25	from using AI outweigh the
		negatives in terms of training
00:04:08;27	00:04:11;24	and maintaining systems.
00:04:11;26	00:04:14;29	The next principle is that
		respect for human dignity
00:04:15;01	00:04:19;04	and privacy must guide AI
		throughout its lifecycle,
00:04:19;06	00:04:22;00	including safeguards
		against discrimination
00:04:22;02	00:04:24;17	and ensuring human oversight.
00:04:24;19	00:04:28;00	Other than designing algorithmic
		safeguards against cyber-attack
00:04:28;02	00:04:30;17	or bias and to enhance
		user control,
00:04:30;19	00:04:35;22	this means establishing clear
		data ownership agreements.
00:04:35;24	00:04:38;05	Stakeholders should
		have meaningful access
00:04:38;07	00:04:41;25	to how AI decisions are
		made, from inputs to outputs.
00:04:41;27	00:04:43;28	They should be able
		to access information
00:04:44;00	00:04:47;09	on the algorithm's design
		and challenge outputs.
00:04:47;11	00:04:49;17	A farmer should have
		the opportunity
00:04:49;19	00:04:52;26	to know why a deep-learning
		model makes the recommendations
00:04:52;28	00:04:56;07	it does or contest legal
		or economic decisions
00:04:56;09	00:05:00;05	that affect them that
		were made using AI.
00:05:00;07	00:05:04;18	AI should perform reliably
		in any scenario, even misuse,
00:05:04;20	00:05:06;27	with clear mechanisms
		for overriding
00:05:06;29	00:05:09;17	or decommissioning,
		if necessary.
00:05:09;19	00:05:12;22	For example, an autonomous
		harvester, like the one
00:05:12;24	00:05:14;11	in the image, should be designed
00:05:14;13	00:05:17;07	to withstand any
		field conditions.
00:05:17;09	00:05:20;22	It should behave safely,
		even if someone's misusing it
00:05:20;24	00:05:22;27	or using it without permission.
00:05:22;29	00:05:26;06	The user themselves should be
		able to safely stop or shut
00:05:26;08	00:05:28;05	down the machine,
		whether it be contained
00:05:28;07	00:05:29;23	within a physical robot
00:05:29;25	00:05:33;03	or existing only in
		the virtual realm.
00:05:33;05	00:05:36;12	The last principle involves
		ensuring traceability
00:05:36;14	00:05:40;26	and systematic risk management,
		allowing for responsibility
00:05:40;28	00:05:44;26	to be clearly assigned
		along the AI's lifecycle.
00:05:44;28	00:05:48;04	That means clearly assigning
		liability, among other things,
00:05:48;06	00:05:51;07	and making it clear how
		remediation is afforded,
00:05:51;09	00:05:56;16	if at all, if a machine
		causes any form of harm.
00:05:56;18	00:05:58;27	The Organization for
		Economic Cooperation
00:05:58;29	00:06:01;28	and Development makes several
		specific recommendations
00:06:02;00	00:06:05;00	for AI policymakers
		and developers based
00:06:05;02	00:06:07;28	on these ethical principles.
00:06:08;00	00:06:11;18	They recommend public-private
		partnerships to drive innovation
00:06:11;20	00:06:15;23	in trustworthy AI, inclusive
		of technical, social,
00:06:15;25	00:06:18;01	and ethical dimensions.
00:06:18;03	00:06:21;14	They further recommend using
		open and representative datasets
00:06:21;16	00:06:24;25	to reduce bias, ensuring
		greater sharing of data,
00:06:24;27	00:06:29;21	and enhancing international
		cooperation and exchange.
00:06:29;23	00:06:33;18	To share open and representative
		datasets, organizations
00:06:33;20	00:06:36;25	and governments need to build
		sharing infrastructures,
00:06:36;27	00:06:41;12	like online platforms,
		to exchange or host data.
00:06:41;14	00:06:44;14	It's recommended that such
		networks promote an inclusive
00:06:44;16	00:06:48;02	digital environment, creating
		pathways for smaller actors
00:06:48;04	00:06:53;05	and developing economies to
		access and use open data.
00:06:53;07	00:06:55;19	Given the rate of
		technological change,
00:06:55;21	00:06:58;00	AI governance should
		be flexible,
00:06:58;02	00:07:03;01	harmonic across national
		borders, for global consistency.
00:07:03;03	00:07:05;16	It's recommended that
		policymakers and governments,
00:07:05;18	00:07:08;18	in particular, invest
		in AI literacy,
00:07:08;20	00:07:10;09	training, and reskilling.
00:07:10;11	00:07:14;11	Furthermore, they recommend
		social protection structures
00:07:14;13	00:07:17;27	to protect against the economic
		consequences of automation
00:07:17;29	00:07:21;13	for displaced or
		replaced workers.
00:07:21;15	00:07:25;05	Of course, ethical guidelines
		are just that, guidelines.
00:07:25;07	00:07:27;21	They aren't legally binding,
		even when dozens of nations,
00:07:27;23	00:07:30;03	including the USA,
		have approved them.
00:07:30;05	00:07:32;26	There are no criminal
		consequences for not adhering
00:07:32;28	00:07:37;01	to such guidelines or
		methods of enforcement.
00:07:37;03	00:07:39;22	But despite there being a
		number of other governmental
00:07:39;24	00:07:42;23	and corporate attempts to
		articulate ethical standards
00:07:42;25	00:07:47;16	for AI across many nations,
		they're all remarkably similar.
00:07:47;18	00:07:50;01	To me, this implies
		that the potential risks
00:07:50;03	00:07:52;08	of AI are commonly
		perceived globally,
00:07:52;10	00:07:54;16	and there is a good
		level agreement
00:07:54;18	00:07:58;29	on what good AI should
		be, at least in theory.
00:07:59;01	00:08:02;04	Agriculture professionals
		perhaps have a unique role
00:08:02;06	00:08:06;15	to play as ethical stewards,
		translating technical standards
00:08:06;17	00:08:08;14	into on-the-ground action.
00:08:08;16	00:08:11;18	As collaborators linking
		farmers with tech developers
00:08:11;20	00:08:14;23	and policymakers, ag
		professionals can advocate
00:08:14;25	00:08:18;29	for ethical AI technologies and
		deployment within agriculture.
00:08:19;01	00:08:22;23	But what might this
		mean in practice?
00:08:22;25	00:08:26;23	Well, look for systems to be
		grounded in fairness, justice,
00:08:26;25	00:08:32;07	and respect for human dignity,
		safety, and sustainability.
00:08:32;09	00:08:35;24	Create inclusive spaces
		for dialogues about AI
00:08:35;26	00:08:39;19	with stakeholders
		of all backgrounds.
00:08:39;21	00:08:43;12	In doing so, you can foster
		trustworthy communication
00:08:43;14	00:08:46;03	and promote AI literacy
		and engagement,
00:08:46;05	00:08:49;22	affording people the resources
		to explore the ethical questions
00:08:49;24	00:08:53;05	of AI deployment in ag.
00:08:53;07	00:08:55;19	By knowing what constitutes
		ethical AI,
00:08:55;21	00:08:57;29	ag professionals can
		review systems on behalf
00:08:58;01	00:09:02;21	of their audiences to help
		guide adoption decision-making.
00:09:02;23	00:09:06;21	The technical and risk aspects
		are important dimensions.
00:09:06;23	00:09:10;26	But is the AI developed and
		deployed in an ethical manner?
00:09:10;28	00:09:13;08	You can help answer
		that question.
00:09:13;10	00:09:16;08	Keep in mind that ethical
		AI, according to most,
00:09:16;10	00:09:20;18	is environmentally, socially,
		and economically sustainable,
00:09:20;20	00:09:24;00	inclusive in its design and
		deployable by a wide range
00:09:24;02	00:09:27;10	of people and operations,
		protective of user rights
00:09:27;12	00:09:31;08	and privacy, safe to use
		and reliable, explainable.
00:09:31;10	00:09:35;01	Decision-making and design
		process is transparent.
00:09:35;03	00:09:39;12	And clearly, someone or some
		entity is explicitly responsible
00:09:39;14	00:09:42;13	for all of those aspects.
00:09:42;15	00:09:48;05	[ Music ]
00:09:48;07	00:09:51;07	Humans have been debating
		ethics for millennia,
00:09:51;09	00:09:54;21	and there's doubtless a lot more
		to talk about in relation to AI
00:09:54;23	00:09:58;00	and ethics that are far beyond
		the scope of this video.
00:09:58;02	00:10:00;18	But I hope this video
		has sparked some thinking
00:10:00;20	00:10:04;06	about what constitutes ethical
		deployment of intelligent,
00:10:04;08	00:10:07;19	autonomous systems in
		agriculture and how you might go
00:10:07;21	00:10:10;20	about addressing these
		issues with your audiences.
00:10:10;22	00:10:14;11	Ultimately, AI affects us
		all, and we all have a stake
00:10:14;13	00:10:16;06	in this ethical debate.
00:10:16;08	00:10:19;13	Thank you very much for your
		attention and your energy.
00:10:19;15	00:10:34;15	[ Music ]